DIP Episode 143 - The Clutch Biostats Review (Comprehensive for all the USMLE exams)

Okay, welcome. My name is Divine. I am a resident and this is episode 143 of the Divine Intervention Podcast. In this podcast I will be essentially given a comprehensive bio-stats review for the USMLE exams. I will say essentially this podcast if you're if you're if you're taking any exam of step one step to CK or step three. I strongly recommend that you make sure you go through this podcast. I'm actually attaching slides and I will highly recommend that you you know download this the slides and maybe I know T them as I go along but this will be a very again like I said very comprehensive review the only thing I really would not talk about in this podcast is bias. Bias is something I think probably deserves its own podcast so I'll probably make a podcast in the future relating to that and just as a pro tip for people taking excuse me for people taking the USMLE exams whenever you get the drug-out questions you should always leave them for last in the block. I'll just say for the great majority of people it usually makes no sense for them to try to do those drug-out questions first. Once you see it just just mark it just mark those questions and move on come back to them later and then the other thing I want to say with regards to bio-stats is do not because I see many people they're just like oh bio-stats is a plug-and-play science and they say oh yeah memorized all the formulas and then you're like man why are you doing so poorly on these bio-stats questions.

The thing is to be perfectly honest it's it's not very common for you to have to do extensive math to arrive at the answers to most USMLE bio-stats questions most of them just kind of depend on having a thorough understanding of what's going on like having a thorough understanding of the concepts usually that gets you all of the way there for the most part with many of these questions because most times they ask you about relationships the ask you about patterns things that are not necessarily abstractible by just memorizing the formulas.

So you will notice today in fact to be to be honest with you and again I'm not saying this from a point of pride I'm just telling you what I do I essentially never set up two by two tables for most bio-stats questions don't get me wrong for some bio-stats questions you know you do need to you know do some math right things like likelihood ratios and stuff like that but for like maybe like 95% of bio-stats questions I do I usually do not do any kind of math or set up like a two by two table or anything like that most times if you kind of like understand the concept and really know like like not just like oh I know the formula like really understand what the concept speaks typically the answers are pretty obvious on on MBME exams right so my goal with this podcast is to teach you how to think about bio-stats I essentially want you to feel like oh bio-stats is pretty easy I actually feel like I have like a very good solid understanding of this and for people that are taking the step three exam especially you better know your bio-stats I will say probably like one out of every like five or six questions you'll meet on the one of your exam is gonna be a bio-stats question so don't blow that stuff off it's a recipe for disaster and then this podcast I will also say it's probably very useful for people that are taking like the internal medicine entry exam and the internal medicine abroad exam because this they tend to test a lot of these concepts relating to like bio-stats bio-stats bio-stats the test them in very unique ways and I think this podcast should also make you very well served to deal with those kinds of problems so without further ado I'm just gonna go ahead and jump into it and I love the way I structure this I have like questions like really like 90% of this presentation is based on questions so again by sort of walking through these questions I really do think

it will give you like really by the end of this podcast I think you should have a very very solid basis for bio-stats the thing I will just say though is this is one of those podcasts you don't want to speed through again I'm saying this as someone that watches many things at 2.5 x speed or 2 x speed sometimes but really for the most part you probably want to speed through this so that you don't lose the friend of thought okay well try to really make sure as you're going through these slides try to make sure that you're really understanding what's going on as you listen to me make annotations that I think at the end you should feel very comfortable with with bio-stats so let's jump right into it so question one so a new serum test is created to screen for peripheral arterial disease the sensitivity of the test is 80% the most accurate interpretation of this statement is so a new serum test created a screen for peripheral arterial disease sensitivity is 80% the most accurate interpretation of this statement is what so option A patients with positive test results have an 80% chance of having the disease option B in patients with negative test results 80% do not have the disease option C in patients who have the disease 20% will have a negative test result and then in option D patients with negative test results have an 80% chance of not having the disease so what do you think the right answer here is our pause if I read you so the right answer here is actually option C right the right answer here is option C right and we kind of think about it right like the thing is again like I said earlier answering in B and B questions for the most part especially the bio-stats questions I mean yeah you can do math to get to the answers but most times you you don't really need math to answer these questions many times you just need to think you can usually finnigle your way to the

right answer and sometimes like even with knowing the formulas I see some people they are like how do I even set up the math in the first place if you don't have the understanding it'll be hard for you to set up the math or if you set up the math you set it up wrong okay so just something to keep in mind right so the thing is sensitivity when you're dealing with a concept of sensitivity you essentially trying to answer this question you're essentially asking yourself of all the population with a given disease what percent have positive test results that's literally all sensitivity deals with of all the population on the given disease what percent have positive test results once you figure out that percent you've essentially figured out your sensitivity the thing is the other percent that you don't detect that truly have disease those will obviously be your false negatives right and if you notice the second word in false negative is negative but the word on in front of it is false so that means that if it's a negative that is false that means it is in fact true right it's in fact a positive right so one thing I recommend especially to people that I tear one on one I tell them to consider using like this a second to first word mantra to keep things straight okay and another thing you want to remember here is that when a test is highly sensitive it tends to have a low false negative rate you cannot see the way that ends much with that so looking at the other answers right so answer choice is wrong right because it says oh patient with positive test results have an 80% chance of having the disease that's essentially describing the positive predictive value which I will get to in a bit option B says in patients with negative test results 80% do not have the disease right again that's essentially expressing the negative predictive value so I'll talk about that in a bit and

then option D says patients with negative test results have an 80% chance of not having the disease again that's just almost like moving the words around for option B okay so again by yourself I promise you it sounds nebulous but it's not hard if you just really try to understand what's going on now the next question so a study is done on a thousand patients with a history of glioblastoma so GBM a new serum test and the name of this test is ST is done to screen for recurrent GBM a hundred patients have a positive ST test and 900 have a negative ST test bring the imaging with biopsy so this is obviously like a gold standard test is done on all these patients and 30 recurrences of GBM are found 10 patients with positive ST tests have GBM and 20 patients with negative ST tests have GBM which of the following best represents the sensitivity of ST tests so I'll just read this real quick again so study thousand patients that have a show GBM new serum test the serum test is ST you do it to screen for recurrent GBM right a hundred patients they have a positive ST test 900 have a negative ST test and then you do like a gold standard test right so like bringing imaging with biopsy and you do it on all these patients and you'll find 30 recurrences of GBM and of those 30 recurrences you notice that 10 patients with positive ST tests have GBM like the GBM recurrence and 20 patients with negative ST tests have the GBM recurrence right and then I'm asking for the sensitivity of the ST test here right so option A is 92% option B is 35% option C is 75% and option D is 50% so what do you think the answer to this is so the best answer here is actually B right so the sensitivity here is actually 33 33% and I'll talk about I guess I'll sort of go ahead and mention it now so the thing is you may say but divine there's no 33% as an answer choice so here's the deal right so don't blame me bl

ame the MBM the thing is your occasionally the MBM actually like does the ST in where they put answers that are not completely exact with the math they just put in exact answers that are kind of close to kind of close to what the real answer is right whatever you see this happening you're like man I've done this math twice I'm still getting the same answer pick the answer that is close as to the answer you spit out from your math okay and again this question sounds really hard and nebulos with all these numbers but again if you do simple math based on an understanding of the concept you'll save the day for you here essentially right so the thing is sensitivity essentially tries to answer the question right again like I said of all the population with a given disease what percent have positive test results right what percent have positive test results right so the total disease population is 30 people right that's me very clear in the question and then the number of those like the the number of those are disease people positive test results is 10 so like 10 of those people have like the positive ST test so your sensitivity is essentially 10 over 30 which is 33% again you can go ahead and plug this into a 2 by 2 table if you want but again I don't think that's absolutely necessary if you just understand the concept this is essentially what I do like one like on my own on exams and the answer here is 3% okay so let's jump on to the next question right so we have a new serum test for GBM okay has a specific a specificity of 90% the most accurate interpretation of this statement is what so option A says 90% of patients with GBM have positive test results option B says 10% of patients with GBM am missed by this test option C says 10% of patients without GBM have positive test results option D says 90% of patients without GBM have positive test results right so let you move

over that for a second so what do you think the answer is here so the best answer here is actually C okay again simple math plus understanding is the way to go with bio stats and really just in general with any subject on the USM at least understanding is usually the way to go right so again kind of like I talked about with sensitivity specificity essentially answers this one question this is how you should try to interpret sensitivity of all the population without a given disease what percent have negative test results again of all population of all the population without disease what percent have native test results that's literally all you need to worry about right so in the question I tell you that the specificity of this test is like 90% right so that means of all the people without GBM 90% tested negative right so that means there's this 10% that should have tested negative but they ultimately ended up testing positive essentially these people are false positives right so that's why the right answer here is C because option C says oh 10% of patients without GBM have positive test results right those are your your false positives okay and kind of like I said that oh a highly sensitive test has a low false negative rate right you see the ends kind of matched there a highly specific test on the flip side also has a low false positive rate you can see how I kind of highlighted both peaks in the in the in the Power Point okay so let's take a quick like sidebar here right you've probably heard of this spin and snout principle right so let's talk about it right so the thing is if you have a test that's like you know super sensitive like a very highly sensitive test people that have disease just sort of walk through walk with me through my thought process here if a test is highly sensitive it means that people with disease should have a positive test result that's what y

ou'd expect right on the other hand if a test being highly sensitive like comes out with like a negative result so if the test results says oh negative then that means the disease should be absent right because again this is kind of building up on things I've sort of talked about already right if the test is negative the disease should be absent that essentially tells you like essentially you're dealing with a low false negative rate right so a negative test should help you roll out disease that's where the snout principle comes from that if a test is highly sensitive that's the sn right it should help you roll out disease right but if you look at things from the other perspective right if a test is like highly sensitive then that means people without disease should have a negative test result right now the thing is if you are like oh okay this test is you know very specific people without disease negative test result if you get a result that is positive from this highly specific test then that means that disease should be present because this highly specific test has a low false positive rate right so positive test should help you roll in disease that's where the spin principle comes from so for a very specific test that's the S if positive that's the P should roll in disease okay on like the snout principle where the S is a highly sensitive test that's the S if negative that's the n should help you roll out disease and this kind of leads into my next point of like screening versus like confirmatory tests the thing is again if you have a test that has very high sensitivity like I said earlier again you may see divine repeating yourself over and over again repetition always helps trust me so in test with high sensitivity right people that have disease you should have positive test results right so a high sensitivity test it actually makes a very good screening test ri

ght because you don't want to inadvertently like miss out on people that have disease right so take for example if you know you're screening for HIV you want a test that is very high-sensitive right so that you don't miss out on anyone that has HIV right because imagine what will happen if you miss out on person that truly has HIV be going into the world have sex with other people give those other people HIV right that's that that's not good right so you want to try to prevent that that's why you use high sensitivity tests for screening right that's why essentially for HIV we use the Eliza test the Eliza test is extremely sensitive for the detection of HIV but if you have a test with very high specificity people without disease should have negative test results right that's what you'll expect so the thing is these high specificity tests they actually awesome at confirmation they are very good confirmatory tests right because the thing is you don't want to like wrongly label people without disease as having disease right so the thing is these tests that are again very highly specific they're good at labeling people without disease so that if you get like you're like oh this test is very specific it's a high-specific visited test if you get a positive result right and you know by definition already have how much this point home over and over again you know by definition that a high specificity high specificity test has like a very low force positive rate if you get a positive result from a very specific test then that patient very likely has the disease right so this is why in hitch like if if a person has a positive Eliza test for HIV you don't like run out into the room and tell the patient oh you've got HIV well you should probably see it in a more patient centered fashion maybe a more subtle tone of voice but that's a different story for another day but I really do

the Eliza test you don't tell your patients the results just yet right you the next thing you go you do is you tell the you you do a Western blood right to confirm right the Western blood you do them after a positive Eliza right because again there is nothing that feels as terrible as they're in a patient oh I'm sorry you have HIV and then you go and do the Western blood and you're like oh crap this patient doesn't have HIV right you'd no want to do that that kind of looks like a like a lawsuit we didn't to happen so you don't do that right so although I guess with that said remember the Western blood is not is no longer what we do in most places to confirm a HIV at least in the US okay so let's jump to the next question right so question four so which of the following points best represents the region of the graph with the highest positive predictive value for the detection of type two diabetes melodies so I'll encourage you to look at the graph and then tell me what do you think what is the region on the graph that represents the highest positive predictive value for the detection of type two diabetes okay so the right answer here is actually D right so the thing is these questions they're like super super super annoying right well let me just essentially tell you the trick that you should follow right the thing is you should always try to have a frame of reference we have a frame of reference these questions are cake right so here's what you should always think about the region that has the highest positive predictive value will be the region that has the highest specificity okay and that will essentially correspond to the region where you do not miss a single person that does not have disease or pick that again the region of the graph that has the highest positive predictive value is the region that has the highest specificity you see how the piece match the PMPPV

the PIN specificity and I mean think about it right you say that a test has like for example 100% specificity if all the people that don't have disease get a negative test so you're essentially looking for the region on the graph where you do not miss out on anybody that has a negative test and if you look at this graph right on slide 10 right you see that there's the part of people that the part of people that that don't have disease so the right answer here should be C because again if you look at this graph right you notice that the blue like the blue line sort of tells you okay these are all the people that do not have disease right they don't have any disease right and if you look at a like B if you look at B B the point B on the graph you know it has most of the people that don't have disease but it doesn't have all of those people but if you look at C everything to the left of C includes everyone that does not have disease right so that is the region of the graph where you have the highest specificity so that's the region of the graph where you should have the highest positive predictive value okay that is essentially the region of the graph where you absolutely do not miss out on anyone that does not have disease okay again remember positive predictive value essentially means the percent of people with positive tests that have disease okay so positive predictive value means the percent of people right so you it's like it's basically like the question you answer when a patient like me to and says like let's say you you give a positive test result to a patient if the patient asks you what doc how sure are you that this result is correct essentially that patient is asking you about your positive about the positive predictive value of the test like oh okay of all the blue that have positive test results what percent of them are correct what which of them actually

are like real true positives right if you see that you're dealing with a positive predictive value okay so again very very high yield to make sure you understand these things and kind of like a quick sidebar so that you don't again mix this up on exams let me try to buttress this point the sensitivity of a test right I said this earlier represents the percent of people with the disease that have positive test results the sensitivity of a test is the percent of everyone that has disease that you get tested they get positive test results positive predictive value of a test is the percent of people that have positive test results that have disease okay that's the percent of people with positive test results that have disease okay do not mix this up percent of people positive test results that have disease so again please please please don't mix this up like if you're if you switch the words like before and after like who have you essentially get the right definitions for both of these states okay or please please please I'm begging you try not to mix this up on an exam so let's jump to the next question so question number five so which of the following points best represents the region of the graph with the highest negative predictive value for the detection of type 2 diabetes militaries right so which of the following points best represents region of the graph with the highest NPV for detecting type 2 DM so the answer here is is B okay the answer here is B right so remember again this question is super annoying but again not very hard right because essentially the the thing you're looking for on the graph is the region with the highest NPV right is the region that will have the highest sensitivity right is the region that will have the highest sensitivity kind of see how the ends match right so that's essentially the region of the graph that does not miss anyone with d

isease right once you remember this you're pretty much you're pretty much set right so if you look at this graph right you see the red line kind of shows all the people that have disease if you chose C yeah you will you'll miss very few people that have disease but you're still you're missing some people won't like the part of the graph where you don't miss anybody at all that has a disease that's where option B comes in because everything to the right of option B those people all have disease I really hope this this makes sense right again remember NPV of all the is essentially like the percent of people negative tests right that do not have disease okay those are things again you want to keep in mind then again you see this next sidebar again please don't mix this up right if you notice I'm kind of like trying to drop parallels and keep things as uniform as possible right specificity I already talked about this earlier is the percent of people without disease that have a negative test result that's it right what NPV is different NPV is the percent of people with negative test results that don't have disease okay those are two things they sound very similar but they are very different and you can already begin to opine that your friends at the mbme want to write questions with the mix and match those words around to try to mess with your head do not fall for that remember divine saying do not fall for that okay make sure you really understand what these terms mean so let's jump to the next question right so we have a clinical trial is conducted to measure the effectiveness of the I am test so is a test I'm calling it I am test as a screening tool for the detection of testicular cancer so 500 I am tests are obtained 20 men have positive I am tests and are found by testicular biopsy to have testicular cancer 180 men have positive I am tests and a negative for testicula

r cancer by biopsy 290 men have negative I am tests and a negative for testicular cancer by biopsy 10 men have negative I am tests and are found to be testicular cancer positive by biopsy what is the negative predictive value of this test for the detection of testicular cancer so let you move over that for a second and then we'll talk about the answer so the correct answer here is actually a right the correct answer is a right so again you see you see people they see these questions tons of numbers they begin to hyperventilate you do not need to do that again the key thing is ask yourself what is the quantity that's being tested in this question it's a negative predictive value and then the next thing is you define it like you use like the logical definition you have you sort of write that out and then say okay so what are the numbers I need to fulfill this definition I just wrote out right you bring out the numbers you need and many times this should get you to to your answer right so notice there are all these numbers in the question right but look at like again I can already imagine people like setting up these two by two tables again you don't need to do any of that crap right you you really can reason through these things and make your life a lot easier and go through questions a lot faster right so if you look at it like I said earlier negative predictive value is of the people that have negative tests what percent don't have disease that's essentially all you're trying to do the percent of people with negative test results that don't have disease so all you need to do is ask yourself oh okay what the people that had negative test results well there's 300 of them so that's in your denominator right and then of those people that had negative test results how many of them do not have testicular cancer it's 200 and 90 of those people right it's 290 because the ques

tion clearly says 290 men have negative IM tests but they are negative for testicular cancer by biopsy biopsy is like a gold standard test here right so you essentially just do like 293 and that's 97% and that's it notice I did not set up any two by two table anything like that and you see divine no that's not how it works on exams I promise you if you really dedicate yourself to understanding these things this is how these questions work on NV Me exams so let's jump to the next one so if the cutoff for a positive IM test result for the detection of testicular cancer is 5 which of the following best represents the outcome of adjusting the test cutoff value to 1 right so the outcome for a positive IM test result for the detection of testicular cancer is 5 which of the following best represents the outcome of adjusting the test cutoff value to 1 so option A says PPV would increase so positive predictive value so PPV would increase but MPV would decrease option B says specificity would decrease but sensitivity would increase option C says PPV and MPV would both increase option D says sensitivity and specificity would both increase so let's you all over that for a second you can pause try to work it out okay so let's talk about this right so again like I said define the quantity that's being tested right and then essentially like come up with your own answer in your mind first before you then start looking at the answers because the thing is I see many people make this mistake on exams and it's usually like it's essentially not a great way to take tests in general but looking at the answer choices first before you look at a question it just begins to mess your mind up it begins to sway your mind in like every which direction and then you end up getting confused right and then you start making assumptions when you're reading the question so this way you need to be careful r

ight like analyze what is being tested in the question and then you go after your answer right so look at this question for example like the prior cutoff is 5 right so it's like oh if you have a if you have an I am test valuable 5 you have testicular cancer right think about it if you bring it down to 1 right you're essentially being like you're essentially setting yourself up for a situation where you would catch every single person as the circular cancer it's like if whatever love value just rises a smidge just goes like above one by any stretch you're like boom you have testicular cancer so the thing is you'll essentially not miss anyone with testicular cancer so if you think about it if you're not missing out on anyone that has disease that means your sensitivity must be going up right your sensitivity has to be going up and the thing is whenever your sensitivity is going up then you already know right off the bat that your NPV will go up already talked about this earlier remember the ends match as your sensitivity is going up your NPV is going up right and the thing is as your sensitivity is going up believe it or not your specificity comes down okay so your sensitivity and specificity typically going opposing directions although that's not always true what that's beyond the scope of our discussion here today and then PPV and NPV the typically also going opposing directions okay so the right answer to this question is actually be right because the specificity is decreasing but the sensitivity is actually going up because you're essentially not missing out on anyone that truly has anyone that has let me not use the word truly you're not missing out on anyone that has a testicular cancer yes you may say divine but by learning to one you know yes you're catching everyone that has the circular cancer but you're also like catching a lot of people that a lot of people

that do not have the circular cancer that is absolutely right that is absolutely right so the thing is whenever you try to make a test like super sensitive you also also also get a lot of a lot of you also get a lot of false positives right you also get a lot of false positives so I know some people may already begin to get a little confused with what I just said remember what I said earlier I said that if a test is very sensitive if you have a high sensitivity test it has a very low false negative rate okay a high sensitivity test right has a very low false negative rate you would actually extend that knowledge a little bit further and tell yourself that a high sensitivity test has a very low false negative rate but it also has a very high false positive rate if your test that is super sensitive you're also going to catch a lot of people that the test calls them positive but they actually do not have disease okay so that's kind of like the trade off for getting a very high sensitivity test that is why whenever you do a screening test that is highly sensitive you also want to do a confirmatory test that is highly specific so that you can weed out that big number of false positives that you're getting from being able to catch every single person that has disease because a highly specific test has a very low false positive rate so you're essentially using the low false positive rate benefit of a high specificity test to counter the high false positive rate side effect that you get with a high sensitivity test I really hope this makes sense again I'm taking my time because bio stats is like a source spot for many people so I want to make sure that you're essentially an expert by the time you're done with this with this podcast okay so now let's jump to the next one right so a medical student at Johns Hopkins that's that was actually my alma mater for medical school so me

dical student at Johns Hopkins invents a drug that improves survival in patients with glialblastoma multi-formy right so GBM by seven years right which of the following changes will be seen a few years after after the drug is approved by the FDA right so option A the sensitivity of screening tests for detecting GBM would decrease option B the prevalence of GBM with increasing the population option C the positive predictive value of GBM detection tests would decrease option D the incidence of GBM would increase in the population option E the specificity of screening tests for detecting GBM would increase and then option F says the negative predictive value of GBM detection tests would increase right so I'll give you some time to sort of think about that so pause the pause the recording and then try to come up with an answer so let's walk through this right so we we invented this drug right that improves survival in patients with GBM so it doesn't cure so really a question is carefully on the exam I'm not saying that we're carrying anything no we're just saying that it improves survival in patients with glialblastoma by seven years I mean as some of you already know GBM is a horrible disease right like most people that get GBM the median survival is like 14 months is like the worst possible kind of brain cancer person can have right so the thing is you've invented this drug by essentially inventing this drug that improves survival you'll effectively keep more people that have already been diagnosed with GBM in life right you're just keep like they've been diagnosed with GBM but they're able to live longer because of this drug so the thing is because people that you know they've been diagnosed with GBM you're giving them around for longer the prevalence of GBM in the population would increase right the prevalence would infact go on right because if you think about it eve

n if you get like a new person that gets diagnosed with GBM let's say oh ideally they would die with like ordinarily I'm not saying I did you I never want anyone to die but uh part of my English there right but like let's say usually people that get GBM dying six months right what if you give them a drug and then they're living like seven years or they're about then that tells you that okay like the number of people with GBM will begin to accumulate because yes you are not changing the number of new people like it's not like oh by giving this drug you are reducing the incidence of GBM now people will still get GBM at the same rate but when they get GBM they live longer than they used to live before so the prevalence actually goes up right and the thing is as prevalence goes up the positive predictive value actually goes up as well okay so as prevalence goes up the PPV actually goes up so that's why option C is actually wrong right and the thing is remember I just told you in the previous question that PPV and NPV are inversely related to each other as the PPV goes up the NPV should go down okay so that's why option F is wrong right and again remember when you change the prevalence of a test it doesn't really do anything to the sensitivity or specificity of that test right so that's why options in here you know flat out wrong really the only things that can change the sensitivity or specificity of a test is if you like change the actual test so let's say like the previous question I think it was like a question or two ago we're actually like change the color value for the test okay so again you still be diagnosed in GBM at the same rate so the incidence really does stay the same and kind of like as a sidebar right again you may ask like divine why does the positive predictive value increase when the prevalence increases well stick with me here for a second right the th

ing is let me just give you a simple scenario that should clear the sub pretty easily for you so think of it if you have a patient you know they come into the eating December and they have you know like fever and rhinitis so like running nose and they have like my algae so like everywhere hurts right they likely have the flu right the thing is if you did a flu swab on these people and it came back as negative would you really believe those results you probably would not try because you're like come on man this thing looks like classic classic classic influenza right so the thing is the prevalence of the flu goes up in December right so if a test result is negative you don't believe it that's like an easy way to remember that the negative predictive value actually goes down right but if the test is positive you're like oh yeah yeah I mean I kind of knew this before I even did the test that you for sure have the flu right so you are more likely to believe the results of a positive test right you are more likely to believe the results of a positive test right so that's the thing you are if you're taking it another way like you're less likely to believe the results of a negative test during a high prevalence period for a disease okay so again stated another way right if a disease is common you are more likely to believe that the result of the test is positive okay again I'm repeating things different ways but again I just want you to feel super grounded in bio stats by the end of this podcast and again as I kind of alluded to in the question right incidents it's the number of new cases of a disease that you know have been diagnosed within a specific time period so new cases new cases that's the name of the game right and then prevalence is essentially the number of people that are alive at a given time period so you define the time period and say oh how many people are al

ive that have a certain disease right at that time period that's the that's the prevalence so let's go ahead and jump to the next question okay so question 9 so an M2 so like a second-year-old student so an M2 researcher at the Gifted Medical Student Institute plans to study the effects of consuming high amounts of care on the development of fiochromositoma he plans to publish the results of his study prior to graduation which of the following study designs presents the most appropriate means of completing the study so option A says around the mice control trial option B says a prospective cohort study option C says a crossover study option D says a case control study and option E says a case report so pause try to call up with an answer okay so let's let's walk through this right so if you think about this the phenomenon that this researcher is trying to you know work on is super super super right it's super rare I mean like fios as common as fios are an exams they actually know that common in the real world I mean like I remember like at most of these major medical centers they may see like one a year or one in two three years right so feels a very uncommon okay and the thing is this is a second year medicine so he probably has like you know like two three years before he graduates so the thing is if this student said oh you know what I want to you know do the high impact study do like a prospective cohort study or do like a randomized control trial the thing is it will literally take this patient like 60 plus years to get like me if not more to get results of from this study because again fios are super rare and chances are before you find a person that consumes scale and you're able to then establish that oh you know what yeah this scale really played a big role in this person getting a few cromoseidoma again that would literally take like a ton of time right that

this much student does not have so the thing is the concept here you want to understand is if you're trying to study like like quantities that are rare like like ref phenomena you almost always want to do case control studies on mb and exams okay because the thing is when you then get some really you essentially like you know you take people that have a feel chromosome to my you take people that don't have feel chromosome to my but are very like similar characteristics right so like they have similar ages similar like environment they lived in like similar comorbidities and all those things and then you ask them you look back to the exposure to kill in the past right the thing is when you you know do this study you can then use that as the basis for like a more detailed study like around the mice control trial or like a cohort study right the thing is many times doing case control studies provides the impetus for you to generate a new research question that you can then pursue a more formal a more rigorous study design like around the mice control trial or like a prospective or cohort study so the best answer to this question is actually option D so again case control studies again you need to groups right the and they need to have similar characteristics right so group one they should have the disease in question group two should not have the disease right and then you just go back in time and ask them about the exposures right the thing is you can already begin to imagine that there will be a lot of recall bias right with these are case control studies and again you want to make sure that you remember and I'll talk about this later you want to remember that the data you get from case control studies for the most part odds are odds ratios I'll talk about odds ratios in a bit so let's jump to the next question right so we have a professor and two medical students tha

t undertake a case control study over the course of a year and publish their results in a high impact journal so let's say like New England journal or jam or whatever right now which of the following best represents an example of a possible conclusion from their study okay which of the following best represents an example of a possible conclusion from their study so option E says the lock setting decreases pain scores in patients with fibromyalgia okay option B says a combination of so force bovier and le de pas vieux here's hepsy with high fidelity option C says as best of exposure causes mesotheliuma option D says or sole dial administration improves survival in patients with primary biliary colonjitis so if I read all pause and try to come up with an answer so let's talk about this right so if you notice so the right answer here is option C I'll just see that right off the bat the thing is if you think about it right like the researchers they essentially like to look at people with mesotheliuma like if you look at an option C they essentially two people with mesotheliuma and then compare them to people without mesotheliuma right and then they likely you know determine that you know what oh they're all this pool that's having mesotheliuma they seem to have like a history of asbestos exposure maybe they worked in shipyards or whatever right and then based on that you can then say okay let me do like a prospective cohort study of some sort right to follow people that work in shipyards to see if they have like an increased incidence for the development of of of mesotheliuma right the thing is you may see but divine why are options AB and D wrong the thing is those other answer choices are wrong because they essentially involve interventions right I mean look at this for a second option is as the lock setting decreases pain scores in patients with fibromyalgia well how

did you determine that you must have given patients the deloxetine and then measured something measured pain scores okay or like combining so force bovian le dipas wear cures hep C well the thing is you must have combined you must have given those people have not given them hep C that'd be weird right you would have given those people those drugs and then determine that oh like did some blood is and determine oh yeah these people have actually cleared their cleared their viral load right also dial same deal right so those options AB and D all involve interventions right so chances are they are likely wrong right so again notice this is a question that's like oh everyone knows what case control study is right but I'd be willing to opine that a decent number of people may have struggled with this kind of question right so again it's very important you need to try to understand like if there is any one thing you take away from me rambling I mean I would want you to take away bio stats from this but if there's one thing you can take away that would really help you like literally for the rest of your test taking life and even just in patient care is to try to make sure you understand concepts the thing is if you understand concepts you typically will do very well on exams regardless of how difficult the questions may seem okay so that's just something I want to live with you that will really I feel like you can almost like change the course of your test taking life like going forward if you can really like abide by this principle so again kind of like to summarize here right again the case control studies the core studies they do with exposures right would randomize control trials they deal with interventions okay and if you notice I put the word the detour here I just used it to remember to talk about what these studies are right so case control studies right again like I

said I already described them people with disease people without disease right and then that have similar characteristics and then you look back in time and say okay ask them about the exposures see if there any like exposures they were enriched for in the past right a court study is where you know you identify two groups of people one group has a risk factor your group does not have the risk factor and then you just follow them in time to see if they develop a certain outcome right and then you calculate a relative risk and all that stuff from from there right and then around the mice control trial is you take an intervention okay you give it to one group and then you give the intervention to one group and then give like placebo to the other group and then you try to compare outcomes okay so again those are kind of important things to know so let's jump to the next question right so the average normal cd 4 count is a thousand per millimeter cubed of blood with a standard deviation of a hundred per millimeter cubed which of the following best represents the normal percentage of individuals who will be measured to have a cd 4 counted and twelve hundred per millimeter cube of blood option a says 2.51 percent option b says 95 percent option c says 5 percent option d says 16 percent and option e says 68.2 percent so pause try to come up with an answer okay so let's talk about this right so really the big thing you want to realize with this question is that 95 percent of the population right for a population that has a normal distribution they should fall within two standard deviations of the me right so the standard deviation here is a hundred right and the population mean and the mean is like is a thousand right so two standard deviations below the mean that's two times a thousand two times a hundred so two standard deviations below that's 800 two standard deviations abov

e that's 1200 right so that means 95 percent of the population falls within cd 4 count number between 800 per millimeter cubed of blood and 1200 per millimeter cubed of blood right so that means 5 percent must fall outside of this range 5 percent must fall outside of the range so 500 percent must be like between less than 800 like have like less than 800 with regards to cd 4 count upgraded and 1200 right so because we have two groups that can fall under this 5 percent category you essentially need to split that 5 percent in half right so that means there must be two and a half percent that has a cd 4 count less than 800 and then there must be two and a half percent that has a cd 4 count greater than 1200 again make sure I promise you they test this stuff all the time on US similes you better make sure you understand what I just talked about so let's talk about take a quick sidebar here and talk about like p values right so the thing is p values right to essentially use them to express like a probability that oh you know what the results I'm getting from this study I did I entirely by chance okay obviously you want that p value to be as low as possible right because the lower the number is then the more you know confident you feel like oh yeah you know what because think about it right it's like if the results I mean let's take it to an extreme let's say the results of like you're like man the results of this study I did you know the results could have this results could have been obtained by chance like 50 percent of the time and obviously that study is never going to get published right it's gonna get like the one's the person that reads the let's say like the paper reviewer is reading whatever article or whatever you wrote they'll just toss it in the trash within like five minutes right you don't want to do that but compare with a person that says oh you know what there

's like a 1 percent chance that the results I got from this study were obtained by chance right people feel a lot more you know confident with that right because you're like yeah 50 percent chance versus 1 percent chance right that makes you feel a lot better because you're like okay that means there is a 99 percent probability that oh these results were not obtained by chance versus the other one where it's like oh a 50 percent probability where these results were not obtained by chance right obviously you want to go with the higher with the higher probability okay so the thing is the lower your p-value the more confident you are in the results of a test okay so if you have like a p-value of like 0.05 that is not as good as a p-value of 0.01 because the 0.01 p-value means that there is one chance in a hundred that the study results you get them by chance versus the one that's like 0.05 where it's like for one you have like a one in 20 chance where the results of the test you are getting them by like the results you're observing from that study like due to chance okay so again very high yield to to understand those and the thing is if you're not told basically like the p-value should always use an NV Me question is 0.05 is only they give you a different p-value they don't give any p-values you don't say any p-values in the question always use 0.05 because that's what's most commonly used in the scientific or community okay so let's go to question 12 right so question 12 says four separate drop trials are conducted to test the relative effectiveness of four different three beta hydroxy steroid dehydrogenase agonists in reason libido the mean libido levels in the study with confidence intervals are graft below which of the following statements are true okay so you look at the graph you can see on the x-axis we have like four drugs drug 1 2 3 and 4 on the y-axis we have l

ibido right low at the bottom high at the top and we're essentially again comparing average libido levels three months after treatment with these with these different drugs so here are your answer choices right on the next slide right so option E says drug 1 is more effective than drug 2 option B says drugs 3 and 4 assimilate effectiveness option C says drug 4 it's more effective than drug 2 and option D says drug drugs 1 and 4 show similar effectiveness so what do you think the answer here is okay so let's talk about that right so the thing is you kind of want to again realize this like general principle right with visa confidence intervals the thing is whenever you have two confidence intervals cross each other right so if like the aligns overlap then that means there is no significant difference between those two things okay again these scenarios are like super common on the usml exams so really statements AB and D should be correct here right because it means he says drug 1 is more effective than drug 2 if you notice drug 1 causes gives rise to a higher libido like value than drug 2 and your lines do not intersect okay so drug 1 really is in fact more effective than drug 2 option B says drugs 3 and 4 assimilate effectiveness again if you notice drugs 3 and 4 they are lines they are confidence interval lines cross each other right so that tells you that there's no there are no significant differences between drugs 3 and 4 and then option D says drugs 1 and 4 show similar effectiveness again the line stress for drugs 1 so they are no significant differences between both the effectiveness is essentially the same okay now the thing is they can test this in a graphical format but they can also test it by giving you just numbers right the thing is they can set like they can give you like you know confidence intervals they can give you confidence intervals of like relati

ve risks or all the ratios or like absolute risk reductions and things like that whenever you see those things you want to be careful here's the principle whenever you have a and a bio starts quantity that is driven by ratios that's essentially like a ratio right so like odds ratio relative risk the thing is if the confidence interval crosses 1 then you have results that are not significant so the p value for those kinds of studies should be greater than 0.05 on the other hand if you have a quantity that's driven by differences right so like the difference between two means for example right or something like absolute risk reduction the thing is the results will be non significant right so you have like p value greater than 0.05 if the results of your confidence interval crosses 1 so you may see divine why is that the case well here's the thing if for example you have a confidence interval of a relative risk that crosses 1 that means for a relative risk to be 1 it means that the risk like the the risking your numerator divided by the risking your denominator is the same because the only way you can get the number one from dividing two things if it is those two is if those two things are the same right so that's why if you have like a ratio driven by the stats quantity and the confidence interval crosses 1 then that quantity there's no signal there are no significant differences between the two things you're trying to measure right now for a difference right if the confidence interval crosses 0 there are also no significant differences so why is that again walk me here for a second the only way the difference between two two between two numbers ends up being 0 is if those two numbers are the same right it's like you can get 0 from doing 2 minus 3 no but you can get 0 from doing 2 minus 2 right so if the confidence interval for like a difference between two things cross

es 0 it tells you that those two quantities must be the same okay so that's why confidence interval for difference-driven quantities difference-driven by-o-stats quantities when because 0 it tells you that you have results that are not significant so let's go into the next question right so we have a study that is done to assess the relationship between vaping in college and by the way vaping is a terrible idea and I'm sure studies are going to come out relatively soon about people getting like really bad long injury I've certainly seen a lot of that on the news recently okay so studies don't assess the relationship between vaping in college and the future need for long transplant right the study yielded a relative risk of 3.5 with a p value less than 0.05 which of the following represents a possible 95% confidence interval from this study okay so option E says 0.5 to 3.5 option B says 2 to 4.5 option C says 3.5 to 6 option D says 3.9 to 7.1 and then option E says 0.71 to 3.68 so pause and try to come up with an answer okay so let's talk about this right so the thing is the if you look at options in E options in E and E are clearly wrong right because notice the confidence interval for options in E include 1 right but this study is measuring a relative risk right and remember a relative risk is a ratio okay the relative risk is a ratio right so you cannot have like significant results and have the confidence interval question that's just flat out right right if you also look at options in C options in C also wrong because if you notice the relative risk that what we sort of got from the study begins or ends the confidence interval right this is literally not possible right if you ever get like results from a test the confidence interval includes the result of that test but it does not start right otherwise called a confidence interval in the first place right you take

that value you get from the test and then like use your z scores and I'll talk about that in a bit use your z scores to say oh this is the range lower than this number I got from the test this is the range higher than this number I got from the test okay so results that you obtained from a study have to be within a confidence interval they cannot begin or end the confidence interval right and then this clearly wrong again because again it it does not include the value you got from the study right like the value of the study is like a freaking 3.5 you can have a you can have a a confidence interval that is like 3.9 to 7.1 where is the value from your study it's not there it's not within range so it cannot be cannot be correct okay so the right answer here is actually option B okay 2 to 4.5 because that includes the number that includes the the number you get from your from your study okay and the confidence interval does not cross one okay so let's go into the next question so we have a study so studies don't assess the effectiveness of a new drug D for the treatment of GBM right so like glioblastoma multi-formy all patients enrolled in the study received the current standard of care in addition to receiving standard of care group A so this is the experimental group they received drug D group B received standard of care and a sham drug Y so that's essentially like a placebo right of the 40 patients receiving D so drug D it's die over the course of the study on the other hand of the 40 patients receiving Y 20 die over the course of the study so in this question I'm asking what is the number needed to treat for drug D so option A is 2.7 option B is 3.3 option C is 13.3 option D is 5.0 and option E is 15.5 so pause try to come up with an answer okay so let's talk through this right so really if you want everyone to calculate the negative number you need to treat right th

e big thing you want to remember is you essentially want to find the difference in risk between the patients that were exposed to some intervention the patients that were not exposed okay so you take the difference and then the number you get from that difference just divided into one that's all you literally need to do right so if you notice I'm not setting up any two by two table or any of that crap I'm just again just reasoning through the concept and arriving at the answer right so basically you ask yourself what is the risk in people that have the God drug what's the risk of like of death in people that God drug D you take that risk what is the risk of death in people that drug God drug why you take that risk right drug why is the placebo is the sham it's the sham drug right and then you take the difference between those numbers divided into one and boom you're home free right so basically right like the number in energy trade is like one divided by your absolute risk reduction right so it's like again 40 people God drug D eight of them died so like 20% of those people died right 40 people God drug why in addition to standard of care right so the drug D people right also remember they also got standard of care right so but 40 people God drug why and 20 of them died within the pair so that's like 50% right so the risk difference is like 50 minus 20% that's 30% right if you divide 30% is like 0.3 if you divide that into one then you're left with 0.33 right you're left with a 0.33 right I mean sorry one divided by 0.3 that's 3.3 oops sorry about that that's 3.3 right and that's the right answer so the answer here is B right and if you understand the number in energy trade you know the number in it to harm right it's essentially like the same calculation right but the thing that happens here is that like the rate of harm in the person that gets the drug like the inte

rvention tends to be higher than the rate of harm in the person that does not get intervention right so let's see the intervention is like getting like I don't know let's say it's like a drug that ultimately gets pulled off the market so let's say like a thousand people study on a drug I mean yeah a thousand like a thousand people get them into a study 500 of them you study on a drug 500 you start on placebo and then like 10% of the people in the group that get the drug die and 1% of the people in the group that do not get the drug die right and that tells you that it's hazardous to take that drug right it's hazardous to take that drug so like die Ethel best still best for example it's probably like a good example of stuff like that right so you may see divine too many formulas to remember now I'm numbering it to harm numbering it to the tree blah blah blah how do I remember this let me actually give you a one-of-a-rule and then your set essentially take one and divide it by the difference between the risks for whatever two groups you're doing right the only thing you just need to make sure you do is try to make sure that you put the higher numbered risk first before the lower numbered risk so because you always want to get like a positive number from doing like your from doing your risk in exposed versus risking on exposed okay again very high yields to understand that so again quickly as a sidebar here right if you want to calculate the relative relative risk essentially you know just take the risk in people that are exposed divided by the risk and the people that are unexposed right so for example right so let's say you have like some kind of cohort study where you're comparing like smokers and non-smokers right and let's say you put 500 you have 500 people in the smoking group 500 people in the non-smoking group right and let's say a hundred people in that smoking

group develop lung cancer what 50 people in the non-smoking group develop lung cancer right essentially 20% of the people in the smoking group block lung cancer 10% of the people in the non-smoking group got lung cancer right so if you take the relative risk you divide the risk you'll essentially say like risking people that were exposed divided by risk and people that were unexposed you get like 20% divided by 10% which is essentially true right so basically smoking gives you like a two-fold increased risk of lung cancer compared with non-smokers okay so can I really hope you understand that relative risk so let's jump to the next question so if the presence of the smurfy care erythrocytes in the urine has a sensitivity of 90% and a specificity of 45% for the detection of IGN Fropathy what is the likelihood ratio of having IGN Fropathy if the patient has the smurfy care erythrocytes detected on your analysis option A 1.35 option B 0.45 option C 4.55 option D 2.33 and option E 1.67 so I'm sure this is probably the question that most people are like I have no idea what's going on here so let's sort of talk about likelihood ratios right so the thing is the occasionally pop up on the usml so you kind of want to know about that right and many students they kind of ask me like many students I typically they ask me divine when do I know when to use the positive likelihood ratio formula versus the negative likelihood ratio formula right so let me give you a rule that will make your life profoundly simple essentially the thing you do here is if the patient has like a positive result from a test okay for patient has a positive result result from a test use the positive likelihood ratio formula okay on the flip side if a patient has a negative test result from a given test okay use the negative likelihood ratio formula right so if you have to calculate here essentially what you

do is since the patient's right if you notice the patient had like positive test results so because they tell you that oh she does the patient actually does have the smurfy carry through sites detected on your analysis right so this is one of those formulas that unfortunately you'd luckily need to commit to memory for the exam right so to get the positive likelihood ratio the thing you essentially do is you take the sensitivity and divide it by one minus specificity okay so for this question right the sensitivity is like 0.9 right you divide that by one minus 0.45 0.45 is the specificity that gives you 1.67 okay so the positive likelihood ratio here is 1.67 okay remember the formula for negative likelihood ratio is like one minus sensitivity divided by a specificity I've usually found it easy to just remember one and then switch it up right like switch up though like bring up the one for the next one and then you get the you're able to build up the equation of build up the equation in your mind okay so let's I guess quick sidebar like what exactly do I mean by likelihood ratios the thing is likelihood ratios I can give like a very lengthy lecture on likelihood ratios we clearly don't have that time because I'm already at like an hour and eight minutes but again this is again if you listen to this podcast you'll be a biochem master by the end right so be here with me I'll try to be done soon there's probably like nine more slides or maybe you know something like that so the thing is likelihood ratios whenever you calculate like a positive likelihood ratio you're essentially telling yourself how much more likely right how much more likely is a phenomenon given a positive test result right how much more likely is a given phenomenon given a positive test result right on the other hand a negative likelihood ratio essentially tells you how much less likely a phenomenon is w

hen when you have a negative test result right so how much sorry how much less likely a phenomenon is when you have a negative test result right so again that it's kind of like if a person comes in there's this example I love to give that let's say they tell you like when you're interviewing for med school that oh a hundred percent of the like fifty no let's not say a hundred percent let's say fifty percent of the students are coming to this med school matching to dirt right and then let's say you come into med school you take step to seek you get it you get you take step when you get it to 80 you get step to seek you get it to 80 right your chances of matching into dirt are a lot higher than okay let me let me put it this way I think I said the first part wrong so let's say you know you're interviewing for med school they tell you that oh if you come to this med school you have a fifty percent chance of matching into dermatology okay you come to this med school fifty percent chance you're going into you're you're you're you're gonna be able to so let's say you apply to the term your chances of matching a fifty percent right and then you come to that med school you take step one step to seek you get to 80s on both right clearly your chances of matching into the with those kinds of scores are much higher than fifty percent they're pretty more like 99 to a hundred percent right assuming you interview well and the other parts of the application are good and you're not like a robot or serial killer or something right so your your your your likelihood ratio as I adjusted like in the positive direction and it's very positive it if you calculate that likelihood ratios for those who probably be like some really high number right because it's almost like certain right that oh you're gonna match into dirt because you got those positive results aka getting to 80s on both USMLE e

xams okay now let's go to question 16 right so in a study examining the relationship between exposure to ketamine and the subsequent development of neutropenia medical records of 300 children were reviewed 100 children who were exposed to ketamine were found to have neutropenia 50 percent who were exposed to ketamine were found to not have neutropenia 80 children who were not exposed to ketamine were found to not have neutropenia and 70 children who were not exposed to ketamine were found to not have neutropenia what is the odds ratio for this study option A 3.29 option B 2.29 option C 5.67 option D 2.23 and option E 7.16 so pause and try to come up with an answer okay so let's talk about it right let's talk about it right so the thing is again you see this question I say oh divine this is a question that will easily take me 5 10 minutes on an MBME exam it does not have to okay if you pay attention to the rule I will give you I will give you like a slightly different formula it's just a different way of thinking about odds ratios you'll make your life supremely easy right you be like many of whenever I see odds ratio questions I can usually get done with them in like 30 seconds to a minute okay so I'll talk you through how I do that so the thing is odds ratio right so what does an odds ratio mean it essentially compares the odds of a person with disease having been exposed to a given risk factor compared to the odds of a control being exposed to that same risk factor right so the thing is my own formula for calculating odds ratio I call it the logical people product divided by the weird people product so I'll say that again the logical people product divided by the weird people product so what in the world do I mean by that so here's what I define as my logic these are the people that are logical right because think about it if you've been exposed to something bad rig

ht you should expect that you get disease from that thing right on the flip side if you were not exposed so let's say like oh if a person smokes you should explain that he should be the ones getting lung cancer if a person does not smoke you should explain that he should be the ones not getting lung cancer right so basically my logical people product are people that you know obey the classic norms that you expect like logical norms right so they are exposed and they are affected or they are unexposed and unaffected you multiply those two numbers together and then for my denominator the weird people product those are people that they're just super odd it's like let's say for example they smoked but they didn't get lung cancer or they did not smoke and they got lung cancer right so like exposed and unaffected and then you multiply that by unexposed but affected right that's your weird people product right so if you do that for this question right you essentially take a hundred a hundred right those are the people that we're exposed to get the kids are exposed to get them in an anguish in tropinia multiplied by multiplied by aegi right so those are the people that they were not exposed to get them in they did not have an intropinia and then you divide that by 70 right 70 represents the people who were didn't get exposed to ketamine but unfortunately they had an intropinia and you multiply that by 50 right 50 represents the people that were exposed to ketamine but they actually did not get intropinia so logical people product divided by where people product and that gives you an answer of 2.29 okay good so let's go on to question number 17 right so the mean block glucose level of a group of 81 medical students was a 170 milligrams per deciliter with a standard deviation of 15 milligrams per deciliter calculated a 95 percent confidence interval and in words interpreter you

r results so let's just sort of talk through this because I don't have multiple choice for this I was kind of running out of time this morning when I was making this slide so I started you'll see at the end I just summarized some high-yield concepts that I just got really tired of making questions from from them because really making these slides and making this podcast certainly takes a very like a large larger amounts of time that people realize but that's a different story I enjoyed doing it it's my passion it's what I it's what I it's one of the things I live for every day I love teaching people teaching I've done teaching for more than a decade I just absolutely love it I mean personally for me teaching is the is the best job in the world I don't know anyone can ever hate can ever hate teaching but anyone that's a that's a conversation for a different day so let's walk through this right on the next slide right so the mean right the average in this question is 170 makes per deciliter right the thing is if you want to calculate confidence interval so you need to calculate the standard error of the mean first and the thing is to calculate standard error of the mean you typically take the standard deviation and divided by the square root of your sample size okay so you essentially take 15 right and divided by the square root of 81 which is our sample size and we get 1.67 okay now the thing is if you want to calculate standard if z scores right I mean if you want to calculate confidence intervals you need z scores right and we're dealing with the 95% confidence interval here so the z score for a 95 percent confidence interval is like two again if you're like oh like super particularly it's 1.96 but again you it doesn't make sense to use 1.96 on any exam use two it's an easier number to multiply by so you take your to get your confidence interval you essentially take

your your mean right so like 170 plus or minus two times your your standard error of the mean which we already calculated as 1.67 right so you already have like 170 plus or minus 3.34 3.34 is 1.67 times two right 1.6 times two is 3.2 0.7 times two is 0.14 right so 3.2 plus 0.14 is 3.34 that's how I do a lot of my mental my mental math right so essentially you have like 170 minus 1 minus 3.34 or 170 plus 3.34 so your range is like 166.66 to 173.34 right so how do you interpret this and again believe it or not the mbme has been known to put questions that test confidence interval and they essentially put interpretations of the confidence interval as answer choices and then they mix a much things and then you have to pick out the right interpretation right so for this study you can essentially say like 95% confidence that the real mean blood pressure of the medical student population right false somewhere between 166.6 and 173.34 okay another way you can say this is that oh the mean blood pressure for any randomly selected group of 81 medical students will fall somewhere between again 166.66 and 173.34 okay 95% of the time if you essentially repeat the experiment on multiple locations those are just two different ways of expressing the same thing okay again make sure you know to calculate confidence intervals and make sure you can interpret them in in words okay very important to know those states so essentially don't with the questions I have I just want to you know summarize some high-yout concepts that I've just kind of run out of time to make our to make questions out of I love writing questions but writing questions especially writing good high quality questions takes ridiculous amounts of time so I'm just gonna draw quick summary here and then we'll be done right so remember your ROC curves right classically right if you want to pick the best test right so the test

with the highest combination of sensitivity and specificity you want to pick a test that is on the upper most left corner of the graph right and then your cohort studies again I already kind of talked about this you look at two groups of people you have different exposures you follow them into the future for the development of an outcome and you can actually have like a prospective cohort study or you can have one that's retrospective the retrospective one you take two groups of people one got an outcome one did not get an outcome and then you look back to see what kinds of exposures they had right and for some reason our retrospective cohort study looks an awful lot like a case control study okay but usually the case control studies useful like very rare phenomena versus cohort studies especially like the retrospective ones that are used for more common phenomena right the retrospective course studies are more for like common phenomena and it tends to have like less recall bias in comparison with with with case control study although the odds ratio is usually pretty close to the relative risk when the prevalence of a disease is super super super low okay I encourage you to try to reason that out or your own and then remember like if you're going like oh for a normal distribution like oh like one two or three standard deviations about the mean right you want to remember that 68% of the population fall within the within a one standard deviation of the mean 95% fall within two standard deviations right and then 99.7% fall within three standard deviations right so again how you to know those and then if you essentially want to compare two groups of two groups of people with a test right I want to do like the t test remember t for two the team t test for the team two right but if you compare more than two groups you want to use something called the F test sometimes they

call it the annover test on enviames right and then you want to also remember the differences between type one and type two errors right when you incorrectly reject the null hypothesis so let's say the null hypothesis is actually true what you actually end up rejecting it that's a type one error that's an offer error but on the flip side if you incorrectly accept the null so it's like let's say like oh like you do a study and you say like oh um smoking does not cause lung cancer that is not true or that is false right so you're seeing smoking does not cause lung cancer you're accepting the null hypothesis and you are incorrectly accepting it that's a type two error okay that's a better error and remember that uh power is one minus beta right then beta errors have to do with power and I'll talk about that in a in a in a in a in a second right now the thing is whenever you have tight confidence intervals your study is a lot more precise when you have tight confidence intervals your study is a lot more precise but you should feel a lot less confidence though with the results of that study right because you essentially have very little room for error because think about it right if for example pressing offers you say oh you know what um do you want to take this bet this bet pays out a million dollars and in this bet um uh you're betting that the temperature of today is going to fall somewhere between 101.1 and 101.3 you probably would not want to take that bet because it's like there's there's a much larger range of temperatures that fall outside those parameters right so you likely get screwed first of all that tells you oh you know what I'm going to pay your family on if um if you can predict the temperature for today and um if it falls between like 50 degrees and 150 degrees I'll pay out a million obviously you'll take that bet without thinking twice because you know y

our chances of success are a lot higher right so uh you feel a lot less confident in the results of a study when you have very narrow confident intervals I mean confidence not confident confidence intervals okay now how can you increase power this thing is this light is florida high to know for the USML step one you're taking in not just step one you're taking any of the USM Ls and you don't know this concept on this slide you're essentially shooting yourself in the foot right so let's talk about different means of increasing power right so one thing you can do is you can just recruit more people for a study right the more people you recruit the more you more closely approximate the the population you're trying to study right and the thing is if if you really do think about it if you have um if you recruit more people for a study if the effect size is small you'll still be able to capture it because there's just so many people I kind of think of it as like you like taking a dollar from every single person that walks by uh by a street that you leave on right almost like a tour route tour roads are profitable because yeah it may be a dollar that they collect on each tour route although that's not exactly from the east coast the east coast they just the gimp people are for money um but that's a different conversation for a different day but if you're getting a dollar from every card that passes a tour than a million cards pass that tour the day you've just made a million dollars right so the more people you recruit for a study the more powerful your study becomes right another way you can increase the power of the study is to essentially have a large difference between the two quantities you're trying to measure right so again you're essentially like having a larger effect size if you have a larger effect size then you don't need a ton of people to you know make that stud

y work right it's like oh let me complete intelligence between two people uh between two groups of people and one group the average uh test score is 99 and then the other group the every test score is 100 the effect size is like one right you likely will not get significant results from that test right even if those people have like you know like a difference in intelligence but the difference like the effect your friends will capture is so minute and you have so few people right that you are not able to get much from that study right so but imagine if you're comparing two groups of people and one group the average score on the test is 25% the other group the average score is 100% there's clearly a large difference I mean there there is an obvious difference in those people right like one group is clearly smart the other group you know they need to study a little harder than the half right so if you have a large effect size that can also increase the power of a study right and then another thing that can increase the power of a study is that if you have like many of your like a lot of your data sort of like plus string around like one central value if you have that it means your study is like your super precise like you're measuring things like really well because think about it if you measure things really well that means you are capturing any differences that exists at a high level right so whenever you have super like precise measurements you're increasing the power you're also increasing the power of your study in fact if you have very accurate measurements you're also increasing the power of your study remember accuracy is very different from precision accuracy means how close are you to the true value precision means for the multiple values you measure like let's say you measure the same thing over and over and over and over again do you get like do you get repr

oducible values right so do you measure something as 1.5 and then you measure it against like 1.49 you measure it again 1.4 you measure it again 1.51 that's a precise study but if for example the true value of that test is 1.8 your testing your your measurements are precise but they're not accurate because you're like way off from the true value but let's say you measure a value it's like 1.79 right and another one is 1.73 another one is 1.85 those values are not precise because they don't you know they don't close around each other really well but they're accurate because they are very close to the true value and again right again the more confident if you are more precise in your measurements chances are you like you made the measurements without leaving anything to chance so your p-values are very low right so whenever you do a study and you have lower p-values that makes your study have more power than one that has higher p-values and then for the next slide right so this I believe the second to the last whoops the third to the last slide right so the this and other things kind of obvious but people for whatever bizarre is intent to get these wrong on the usml's you need to remember that the fact that something is statistically significant right doesn't mean that it's clinically significant right so like it doesn't mean that oh a drug I do my trial my p-value is less than 0.00 far less than 0.05 so I'm gonna bring those drug to market no it doesn't mean that it's a clinically significant right so think about it let's say like oh you do a study right or you you do around the mass control trial and the baseline block pressure of people block pressures or people at the beginning of the trial is like 130 over 80 and then like two years after you've completed the trial they've got in this anti-hypertensive agent their block pressures reduced to 129 over 79 right and le

t's say oh you do like your calculations you do your confidence intervals um you do your p you you get like significant results p-values less than 0.01 that's good that's a statistically significant result but that drug is useless because it's not clinically significant because it's only lowered your systole blood pressure by 1 in a two-year period and a dastole blood pressure by 1 in a two-year period that's a useless drug right and then just again kind of like some basic things remember right mean is the average median is the middle number right um it's essentially like if you have like an odd like an odd number set of data right so let's say you have like seven values in a set you pass three values at the beginning three values at the end the number in the middle is the median and then the mode essentially represents the most frequent quantity in your data set right and then I'll tell you where if you're given a set of numbers they very likely won't do this on an mbm because that's too easy they're given a set of numbers arranged them in ascending or descending order before you start talking about like mean median mode okay and remember that the mean is affected by extreme values in the set okay so let's say a bunch of people let's say like there are 10 people that take the USM list to CK exam and nine of them get like two tens what one person gets like a 300 that's obviously gonna skew the results right that's gonna skew the mean the mean will appear to be higher than it truly is it won't be very representative of the population right because um there's this one person that just scored out of this rock okay so that's something you absolutely want to make sure you keep at the back of your mind for exams and then um going to the second to the last light here right so if you have a normal distribution right you definitely want to remember that the mean median of moda

all the same that's what makes it a normal distribution in the first place right and then remember that just remember this in like almost like ascending alphabetical order so like mean has EA uh precedes median that's ED which precedes mode right like OD right so mean precedes median which precedes mode right that's the alphabetical order if you remember that that will help you remember what obtains with a negatively skewed or positively skewed curve right whenever you have a negatively skewed curve it means like the kind of like the flat portion of the curve is off to the left right you're just again just write your letters out mean median mode and then put a lot of less than signs in between so the mean is less than the median which is less than the mode positively skewed curve flat portion is on the right same thing write out mean median mode or put greater than signs between those so mean is greater than median which is greater than the and then sometimes your friends at the mbm you can give you like you know they can give you like a curve that shows like a bimodo distribution and then you'll give you answer choices that say which of the following phenomena best represents the curve shown above right and then they'll list like a bunch of like random crap like Hodgkin's lymphoma uh something like Birkets lymphoma slone fastest in liters blah blah blah you certainly want to remember at least these two high-old examples of things that are associated with bimodo distributions right so what do I mean by bimodo it means there are two things almost like two peaks there are two things that are sharp commonly so like Hodgkin's lymphoma for example Hodgkin's lymphoma shows up in very young kids and then you don't see it for a long time again and then there's this big uptake when people get older okay that's a bimodo distribution another common one is like suicide right lik

e suicide like is very high in like young people but also very high in old people right it's it's lower in middle age people that's another example of something that follows a bimodo distribution and then like a lead time bias um essentially it's like kind of like thinking it's like erroneous thinking it's like oh um survival has improved uh when in fact like the survival actually improved because you found the disease right I kind of think of it like prostate cancer is a great example it's like oh let's say you start you know you check a 40-year-old guy you find prostate cancer um you detected you treat it and then this 40-year-old guy dies at 85 and you're like oh yes by detecting uh by doing this test of improved survival in prostate cancer no that is not true chances are that guy may have relieved to the age of like 85 with that prostate cancer um without like necessarily dying from it right so don't mistake detecting something early for improved survival that's classic lead time bias again like I said I will talk about bias I will likely make a future podcast that talks about a lead time bias and then we'll go from there so as I do at the end of every podcast um I do offer one on one tutoring for many exams right so step one two CK two CSTEP three um pre-clinical medical exams third year clerkship self exams um if you're a medicine resident um the medicine training exam the medicine board exams I do tutor a ton of people for those um and then if you're college students and you need to do it in like physics, gen-cam, okam, physiology, histology, biochemistry offer tutoring for all those things and then um the longitudinal tutoring that I offer for people studying out med school through the dedicated period and um people that are studying third year through the dedicated period for step two CK um if you're interested in any of those things reach out to me um it's s

omething I've done with a decent number of people and the essentially all of them have been like wildly successful on their USMEL exams and they've also crushed their pre-clinical and third year shelf exams so if that's something you're interested in just send me an email um just in the in the heading just write longitudinal tutoring and again I can try to point you in the right direction essentially what I do is I meet with these people like like once or twice every week throughout their first and second years and as I'm meeting them I'm tutoring them for all their block exams and but at the same time I mean I'm infusing them with step one knowledge and also like giving them like guided practice through like um ambient-style questions that ultimately like when they get to the dedicated periods like they feel very confident and very ready for the exam right same thing with third years they start out and then I um I teach them like stuff that's pertinent to like I prepare them for all their shelf exams and then they get to their dedicated periods they feel like super ready I mean they're people of tutor that I've got into their dedicated periods and they're like hmm I feel very ready for my exam um so if that's something I'm interested in reach out to me um either send me an email um reach out through the website or send me an email actually list the email here divine intervention podcasts with an s at the end at gmail.com and then if you need one-on-one coaching or advice and for like um iraq's applications for med students applying to residency or amca's applications for college students applying to med school uh if you're going to reach out again I have a ton of experience with this stuff um so like mock interviews personal statements um the experiences section of these applications stuff like that writing rec letters I can certainly um um help you out with any of t

hese things again I've done it with tons of people I have a lot of admissions committee experience so take that for what you will okay so I do hope you get something from the spot cast please please please if you're taking any USMLE exam listen to the spot cast I promise you it will it will help you a ton on the exam so I wish you a wonderful rest of the day have a blessed rest of your day and God bless you I'll see you in episode 144 thank you

A negative test result helps rule out disease (Snout Principle).

A positive test result helps rule in disease (Spin Principle).

There is no statistically significant difference between the two groups.

Case-Control Study.

As prevalence increases, PPV increases, and NPV decreases (and vice versa).

$1 / \text{Absolute Risk Reduction}$ (ARR).

How much more likely a phenomenon is given a positive test result.

Sensitivity / (1 - Specificity).

There is no statistically significant difference between the means of the two groups.

High risk of recall bias (relying on memory of past exposures).

Approximately 95%.

Precision refers to reproducibility (measurements cluster closely); Accuracy refers to how close the measurements are to the true value.

How much more likely a phenomenon is given a positive test result.

Sensitivity / (1 - Specificity).

There is no statistically significant difference between the means of the two groups.

High risk of recall bias (relying on memory of past exposures).

Approximately 95%.

Precision refers to reproducibility (measurements cluster closely); Accuracy refers to how close the measurements are to the true value.