Skip to main content
Loading Explore guides…
Effect Modification in Studies · Acedra
Home
Explore
Medicine
Effect Modification in Studies
Sign in to save
Chat
If YouTube asks you to sign in, open the video on YouTube.
Watch on YouTube
Reload player
Transcript
147 segments
Copy
Hide video
0:06
I'M DR. LAURA LEE JOHNSON, THE ASSOCIATE DIRECTOR AT THE FDA I'M ALSO CODIRECTOR OF THIS COURSE, SO THANK YOU FOR COMING. ANY ELECTRONIC DEVICES ON, PLEASE GO AHEAD AND TURN THOSE TO SILENT MODE. THANK YOU. STANDARD DISCLAIMER. AND THE REASON I PUT THE LECTURE NOTES ON EFFECT MODIFICATION AND COULD BE FOUNDING AT THE END OF THE LAST LECTURE I GAVE IS BECAUSE IT BOTHERS ME WHEN PEOPLE ONLY CATEGORY THIS IN EPIDEMIOLOGY AND TALKING ABOUT OBSERVATIONAL STUDIES. WE'RE GOING TO COVER THEM
0:49
TONIGHT BUT KEEP IN MIND THESE ARE PROBLEMS WHETHER YOU HAVE A RANDOMIZED OR NON RANDOMIZED STUDY. SO IF YOU TALK TO A STATISTICIAN THEY'LL BREAK YOUR VARIABLES UP INTO A HANDFUL OF DIFFERENT TYPES. EFFECT MODIFIERS, POTENTIAL CONFOUNDERS AND OTHER THINGS, OTHER COVARIATENS. REALISTICALLY, WE CALL EVERYTHING, ALL THESE XS THAT YOU'LL LEARN TO LOVE BY DECEMBER, WOE CALL THEM COVARIATES. SO EFFECT MODIFICATION IS WHEN YOU HAVE AN INTERACTION. SOME FOLKS CALL THIS SYNERGY.
1:23
THE TRICK HERE, IT COULD BE LARGER OR SMALLER. THE ASSOCIATION BETWEEN THE OUTCOME VARIABLE AND ANOTHER VARIABLE, IN PARTICULAR, THE INTERVENTION IS MODIFIED BY DIFFERENT LEVELS OF A THIRD VARIABLE. WHAT DO I MEAN? WHEN I'M TALKING ABOUT THAT I'M THINKING ABOUT LET'S SAY I HAVE A NEW CARDIOVASCULAR DRUG. AND IF I GIVE THAT DRUG TO MEN IT WILL BEHAVE DIFFERENTLY IN CHANGING THEIR BLOOD PRESSURE. THAT'S EFFECT MODIFICATION. A CLASSIC EXAMPLE IS WHEN WE LOOK AT SMOKING ASBESTOS LUNG CANCER.
2:10
SMOKING ALONE INCREASES YOUR CHANCE OF CANCER BY AMOUNT A. ABBEST TOSS ALONE INCREASES YOUR RISK OF LUNGER, B. IF YOU WERE A SMOKER AND YOU HAD ASBESTOS EXPOSURE, WAS YOUR RISK OF LUNG CANCER T A-PLUS B OR AN AMOUNT HIGH, BECAUSE YOU HAVE THAT SIX EXPOSER MAYBE IT MAKES SMOKING THAT YOU'RE MORE LIKELY TO GET LUNG CANCER, OR PERHAPS HAVING ASBESTOS EXPOSURE ARE PROTECTIVE SO SMOKERS ARE LESS LIKELY TO GET LUNG CANCER. THIS IS THE EFFECT BEHIND MODIFICATION. INSTEAD OF A NUDGE AMOUNT OF
3:02
DEFINITIONS, NEAL PERCENT, IF YOU TYPE IN EPIDEMIOLOGY, NEAL PIERCE CHAPTER, YOU WILL FIND ABOUT 22 DIFFERENT SLIDES WHERE HE GOES THROUGH WHAT IS EFFECT MODIFICATION. DEFINED FOR STATISTICIAN, DIFFERENT DEFINITION INSTEAD OF EXAMPLES FOR PUBLIC HEALTH WORKERS, FOR LAWYERS, BIOLOGISTS, LOTS OF DIFFERENT PROFESSIONS. SO PLEASE, READ NEAL'S CHAPTER. GO THROUGH IT AND FIND THE DEFINITION THAT FITS YOUR PROFESSION AND SEE HOW THAT WORKS FOR YOU. I'M GOING TO MOVE ON TO CONFOUNDING.
3:37
SO CONFOUNDING IS WHEN I HAVE 2 OR MORE VARIABLES. THE TRICK THAT WE'RE GOING TO KEEP TALKING ABOUT TONIGHT AND SOME OF US ALREADY STARTED TALKING ABOUT ON THE DISCUSSION BOARDS FOR THE LAST LECTURE, THEY MAY BE KNOWN OR UNKNOWN TO THE RESEARCHERS. ONE OF THE COMMON THEMES IS THERE IS A LOT OF INFORMATION YOU KNOW, IF YOU KNOW IT, YOU MIGHT BE ABLE TO COLLECT DO IT DATA ON IT. NOT ALWAYS. IF IT'S UNKNOWN YOU'RE PROBABLY NOT COLLECTING THAT DATA. THAT'S YOUR BIGGEST WORRY, WHAT YOU DON'T KNOW.
4:12
THESE 2 VARIABLES OR ADDITIONAL VARIABLES ARE CONFOUNDED WHEN THEIR EFFECTS ON THE COMMON RESPONSE VARIABLE ARE OUTCOME OR MIXED TOGETHER. CONSIDER IF YOU SAW AN ASSOCIATION BETWEEN CARING MATCHES IN YOUR POCKET AND LUNG CANCER. ARE THE MATCHES CAUSING THE LUNG CANCER OR THE ASSOCIATION BETWEEN CARING MATCHES AND LUNG CANCER, PROBABLY SOME UNMEASURED VARIABLE ASSOCIATED WITH BOTH, LIKE CIGARETTE SMOKING. SO YOU HAVE THE MATCHES TO LIGHT THE CIGARETTE. THE MATCHES ARE NOT CAUSING THE CANCER.
5:10
IT'S SMOKING THE CIGARETTE THAT'S CAUSING THE CANCER. THIS IS THE ISSUE WITH CONFOUNDING. IF YOU NEVER MEASURED THEIR SMOKING STATUS BUT MEASURED THE MATCHES, YOU'VE GOT THE WRONG ACTOR. AN EXAMPLE THAT WAS IN THE NEWS FOR MANY YEARS, IT WAS IN THE NEW YORK TIMES , IN 1981, STUDY LINKS COFFEE USE TO PANCREATIC CANCER. SO WE HAVE VERY FEW WAYS TO TREAT PANCREATIC CANCER. EVEN THE MEDS WE HAVE -- METHODS WE HAVE, THEY DON'T INCREASE YOUR LIFE SPOON TO A VERY LONG PERIOD OF TIME.
5:53
THIS CAME OUT OF A NEW ENGLAND JOURNAL ARTICLE IN 1981. THEY WERE TRYING TO FIGURE OUT WHAT WERE SOME OF THE CAUSES. SO PANCREATIC CANCER PROBABLY DOESN'T CAUSE YOU TO START DRINKING COFFEE. WE THINK THIS DIRECTION OF THE ARROW IS PROBABLY CORRECT. COFFEE DRINKING MIGHT CAUSE PANCREATIC CANCER. COFFEE DRINKING AND SMOKING TEND TO GO WITH EACH OTHER. I DRINK COFFEE, I DON'T SMOKE. BUT THIS DOESN'T HAVE TO BE A PERFECT ASSOCIATION. YOU JUST HAVE TO HAVE SOME ASSOCIATION.
6:29
SO IT'S A CONFOUNDING EXAMPLE. A RELATIONSHIP BETWEEN COFFEE AND PANCREATIC CANCER. BUT SMOKING IS THE KNOWN RISK FACTOR FOR PANCREATIC CANCER, HERE. SMOKING IS ASSOCIATED WITH COFFEE DRINKING, VICE VERSA. IT'S NOT DIRECTIONAL. NOT LIKE SMOKING IS THE RESULT OF COFFEE DRINKING. THE PICTURE LOOKS LIKE THIS. SO I KNOW I HAVE A DIRECTION BETWEEN MY SMOKING AND PANCREATIC CANCER. SMOKING LEADS TO PANCREATIC COFFEE. BY THERE IS NO OTHER DIRECTIONALITY HERE. SO WHAT'S CONFOUNDING?
7:12
ASSOCIATION IS OBSERVED BETWEEN COFFEE DRINKING AND PANCREATIC CANCER. THIS OR THE ASSOCIATION A RESULT OF CONFOUNDING BY CIGARETTE SMOKING. MY MEASURING THE DOVEFY I'M -- COFFEE, I'M NOT DOING A GOOD JOB OF MEASURING THE SMOKING. THIS IS THE RELATIONSHIP I'M LOOKING AT. IN FACT, AFTER MANY YEARS, THE DECISION WAS THIS WAS THE ANSWER. COFFEE DRINKING HAS NOTHING TO DO WITH YOUR PAN YI ATTIC CANCER. SO HOW DO YOU HANDLE CONFOUNDING? THE WAY IS YOU IDENTIFY ALL POTENTIAL
7:50
CONFOUNDERS YOU CAN THINK OF. YOU MEASURE THEM. YOU'RE PROBABLY GOING TO MEASURE THEM ENMANY DIFFERENT WAYS. HOW DO I MEASURE SMOKING, CURRENT FORMER NEVER? HOW MANY PACK YEARS YOU HAVE? THERE ARE TONS OF DIFFERENT WAYS TO MEASURE SMOKING. WHAT MOST STATISTICIANS TELL YOU TO DO, DON'T STRATIFY SMOKING STATUS. DON'T STRATIFY, YOU HAVE TO PUT THAT VARIABLE IN YOUR MODEL. WHICH IS SOMETIMES CALLED AN ADJUSTMENT. WE'LL TALK ABOUT THAT IN THE COMING WEEKS. THE BIGGEST PROBLEM IS FEARING THE UNKNOWN.
8:33
NOW, I WILL ALSO SAY THAT THE HARVARD COLLEAGUES THAT WORKED IN THAT 1981 ORIGINAL ARTICLE ON PAN TREE ATTIC CANCER ACTUALLY DID TRY TO CONTROL FOR CIGARETTE USE, BUT PART OF IT WAS THE SPASMING OF THEIR -- SAMPLING OF CASES AND CONTROLS AND THE GORIZATION OF COFFEE WHICH WE'LL TALK MORE ABOUT TONIGHT. THIS NEXT ISSUE COMES UP IN RANDOMIZED TRYINGS AS WELL AS OBSERVATIONAL TRIALS, NOT AS RESIDUAL CONFOUNDING. THIS IS WHEN YOU HAVE THAT POOR MEASURE. HOW DO YOU MEASURE PHYSICAL
9:08
ACTIVITY, FOR EXAMPLE. WHEN WE PUT A CONFOUNDER IN THE MODEL -- I PUT SOME MEASURE. MAYBE HOW MANY STEPS PEOPLE WALK ON AVERAGE PER DAY. SOME OTHER METHOD. I'M NOT REALLY EXPLAINING THE FULL EFFECT OF PHYSICAL ACTIVITY IN THE MODEL. AGAIN, SMOKING IS ANOTHER EXAMPLE. WHAT VARIABLE ARE YOU PUTTING IN? IS IT THE CORRECT WAY TO MODEL SMOKING INSIDE THAT RELATIONSHIP? RANDOMIZATION HELPS PROTECT AGAINST CONFOUNDING. IT DOES NOT PREVENT CONFOUNDING. WHEN DOCTOR JOAQUIN COMES AND
9:51
TALKS ABOUT RANDOMIZATION AND HYPOTHESIS TESTING, PART OF WHAT HE'S GOING TO TALK ABOUT IS WHAT RANDOMIZATION GAPES YOU. IF YOU DON'T RANDOMIZE OR IF YOU RANDOMIZE POORLY, SO YOU DON'T FOLLOW THE PROCESS YOU SET UP, THE PROBLEMS THAT CAN RESULT. BY THIS IS KIND OF LIKE BUYING INSURANCE IN A WAY. IT DOESN'T MEAN SOMETHING BAD WON'T HAPPEN. YOU'RE JUST TRYING TO PREVENT IT AS MUCH AS POSSIBLE. YOU CAN HAVE NON RANDSEM DROP OUT OWNED ATTRITION. YOU HAVE TO HAVE A WAY TO ACCOUNT FOR THAT. WHO
10:32
RANDOMIZATION TELLS YOU, FOR ALL THOSE KNOWN AND UNKNOWN CONFOUNDERS, SOMETHING IS BY CHANCE OR YOU'RE ABLE TO DIG DANDIFYING OUT WHAT'S POSSIBLY CAUSING IT. NOW, A BIG PROBLEM IN SOME OF THE EARLY HIV TRIALS, I USED TO WORK ON, THE PROBLEM WAS PATIENCE WOULD TEST THE SUBSTANCE. AND THEY WOULD DROP OUT IF THEY WEREN'T GETTING THE DRUG THAT THEY WANTED. YOU TRY TO DO A BLINDED TRIAL, THEY WOULD ESSENTIALLY GO TO A COMPOUNDING PHARMACY OR SOMEPLACE ELSE TO HAVE THE SUBSTANCE TESTED.
11:06
OR IN ONE OF MY STUDIES, THEY SAID, WELL, I'M GOING TO GO BUY MORE OF THE SUN STANCE, BECAUSE IT WAS AVAILABLE IN A LOT OF DIFFERENT STORES. THEY WENT AND BOUGHT A LOT OF ISOFLAVINS AND TOOK IT. BECAUSE THEY WANTED TO MAKE SURE THEY WERE GETTING THE ISO FLAVINS, SOMEONE IS SHAKING THEIR HEAD NO. THAT'S RIGHT. DON'T DO THAT. BUT THAT'S ALWAYS PART OF OUR CONCEPTING PROCESS. SOMEONE IS GOING TO DO WITH YOU HOPE IS NOT IN THE STUDY. WHEN SOME PEOPLE ARE AFRAID THEY'RE GOING TO DIE SUCH AS HIV
11:42
DISEASED AND A LOT OF OTHER STUDIES, YOU HAVE TO FIGURE OUT HOW TO BALANCE THIS INFORMATION AND HOW TO HANDLE IT IN THE ANALYSES. SO THIS IS A PICTURE FROM JOHN POWERS LECTURES. HE TALKS ABOUT CONFOUNDING AND EFFECT MODIFICATION SOMETIMES, AND THIS IS A NICE PICTORIAL EXAMPLE. WHERE THE OUTCOME IS KIND OF YOUR Y AXIS. IF YOU HAVE NO CONFOUNDING AND NO EFFECT MODIFICATION, ACROSS THE BASELINE VARIABLE VALUES, KIND OF YOU'RE GOING TO HAVE STANDARD EFFECT SIZE, TESTED
12:24
CONTROL ARMS WILL LOOK THE SAME. THIS IS WHAT YOU HOPE FOR. IF YOU HAVE CONFOUNDING BUT NO EFFECT MODIFICATION, BASELINE VARIABLE, FOR EXAMPLE, MAY BE INCREASING, THE IMPACT YOU EXPECT TO SEE ON THE OUTCOME IS INCREASING BUT THE DISTANCE BETWEEN THE TEST AND CONTROL ARMS IS THE SAME. KIND OF THE IMPACT IS THE SAME REGARDLESS OF WHICH STUDY ARM THEY'RE IN, BUT THE BASELINE VARIABLE LOOKS DIFFERENTIAL. IF YOU HAVE EFFECT MODIFICATION, THE SLOPES OF THE LINES LOOK DIFFERENT.
12:57
IF YOU HAVE CONFOUNDING AND EFFECT MODIFICATION IT'S GOING YOU'LL OVER THE PLACE, RIGHT? BOTH HAVE A SLOPE BUT THEY HAVE DIFFERENT SLOPES. THIS IS WHAT I SPEND MY TIME PLAYING WITH ON SATURDAY NIGHTS, RIGHT? TRYING TO FIGURE OUT IN MY PLACEBO GROUPS, HOW ARE THEY BEHAVING. >> AND THEN IN HIGH TEST GROUPS -- IN MY TEST GROUPS, HOW THEY BEHAVING? IS IT DUE TO ANYTHING ELSE OR DUE TO THE ACTUAL INTERVENTION? SO MOVING ON, TONIGHT WE'RE GOING TO TALK MORE ABOUT EPIDEMIOLOGY.
13:36
BY TEND OF TONIGHT HOPEFULLY YOU CAN IDENTIFY THE COMMON DESIGNS AND UNDERSTAND WHY THEY MAY OR MAY NOT BE USED TO ANSWER A RESEARCH QUESTION. SO IN THE LECTURE AFTER THIS, YOU'RE GOING TO HEAR FROM TWO DIFFERENT PEOPLE. ONE IS JERRY SOCKS, HE'S GOING TO TALK ABOUT CLINICAL RESEARCH FROM A PATIENT'S PERSPECTIVE. AN IMPORTANT ELEMENT TO TALK TO PATIENTS ABOUT IS WHAT TYPE OF STUDY DESIGN ARE THEY WILLING TO BE IN? THERE ARE SOME PEOPLE WILLING TO BE INTERVENED UPON, SOME ARE
14:07
WILLING TO BE RANDOMIZED. THIS ARE OTHER FOLKS THAT DON'T WANT TO PARTICIPATE IN THE STUDIES BUT MAY BE WILLING TO PARTICIPATE IN SOMETHING LIKE A NATURAL HISTORY STUDY. OR MAYBE WILLING TO GIVE SOME OF THEIR DATA FOR OBSERVATIONAL RESEARCH IN GENERAL. STUDY PARTICIPANT COLLECTIONS ALSO GOING TO BE TALKED ABOUT IN THE NEXT LECTURE BY KATHERINE STONEY, INVOLVED IN ALL SORTS OF STUDIES. MAKE SURE TO ASK HER TONS OF QUESTIONS. ALSO BECAUSE SHE'S A FRIEND OF MINE, SO IT'S GOOD TO ASK HER.
14:39
WHEN YOU GET TO NEXT WEEK, AND THE OVERVIEW OF HYPOTHESIS TESTING, DO WANT YOU TO CONSIDER THAT MOST OFF WHAT YOU'RE GOING TO LEARN ABOUT HYPOTHESIS TESTING IS FOR RANDOMIZED TRIALS THAT DOESN'T MEAN IT DOESN'T APPLY TO THESE OBSERVATIONAL OR NON RANDOMIZED TRIALS. SO DR. WAKE. M IS GOING TO GIVE ADDITIONAL EXAMPLES BUT YOU CAN ALWAYS ASK HIM QUESTIONS ABOUT THESE STUDIES. I WANT TO GIVE THANK TO JERRY AT OHRP, HE'LL TALK ABOUT INSTITUTIONAL REVIEW BOTHERS. HE HELPED TO REARRANGE AND EDIT
15:15
THESE SLIDES ABOUT A YEAR OR TWO AGO. WHAT ARE THE DEFINITIONS? WHAT IS THIS EPIDEMIOLOGY THAT I KEEP TALKING ABOUT? THERE IS A SIGN THAT I SAW WHEN I WALKED IN THE BUILDING TODAY. IT WAS LIKE WHY ME? WHY NOW? THAT'S KIND OF WHAT YOU FIGURE OUT IN EPIDEMIOLOGY. EPIDEMIOLOGY IS THE STUDY OF THE DISTRIBUTION AND DETERMINANTS OF DISEASE IN INJURY. WHERE WHETHER IT'S HUMAN, ANIMAL, PLANT, OR ANY OTHER POPULATION. VAGES USE EPIDEMIOLOGY, FOLKS WORKING IN FOREST SERVICES USE EPIDEMIOLOGY.
15:51
EVERY ONE USES EPIDEMIOLOGY. BUT BASIC TENANTS, DISEASE DOES NOT OCCUR AT RANDOM. THERE IS SOME TYPE OF CAUSAL AND PREVENTIVE FACTORS THAT CAN BE IDENTIFIED IF WE DO A SYSTEMATIC INVESTIGATION OF DIFFERENT POPULATIONS OR COMPARING SUBGROUPS OF INDIVIDUALS WITHIN A POPULATION. NOW, WHAT MOST PEOPLE THINK OF WHEN THEY THINK EPI, WHICH IS WHAT PEOPLE SHORTEN EPIDEMIOLOGY TO, THEY THINK EPIDEMIC. SO THIS IS A POPULAR SEVERAL YEARS AGO, CALLED THE HOT PHONE. INTERESTINGLY, BULLET IS BACK ON
16:31
OUR MINDS AGAIN. THEN THERE ARE MOVIES, SO YOU SEE THESE TV SHOWS. IT'S MORE EXCITING THAN WHAT MOST OF MY EPIDEMIOLOGY FRIENDS DO. MOST OF THEM SIT IN FRONT OF A COMPUTER OR EVEN THE ONES IN THIS SERVICE THAT GOES AND HANDLED OUTBREAK, THEY GO AND THEY'RE ON A LOT OF CRUISE SHIPS. PEOPLE GET THE NOROVIRUS. BUT EPIDEMIOLOGY IS MORE THAN STUDYING THESE EPIDEMICS. IT'S ALSO THESE BIG COHORT STUDIES. SO THE NURSES HEALTH STUDY, FOR EXAMPLE. AND THE CARDIOVASCULAR HEALTH DECIDE.
17:08
ACTUALLY A BIG COHORT STUDY THAT SEVERAL OF YOUR CLASSMATES STARTED TALKING ABOUT ONLINE LOOKING AT A EBOLA SURVIVORS, DRYING TO DO A NATURAL HISTORY STUDY. THERE ARE MANY THINGS IN BETWEEN. BUT EPIDEMIOLOGY IS REALLY CONSIDERED BY SOME TO BE THE CORNERSTONE OF PUBLIC HEALTH RESEARCH. ONE OF THE ARTICLES THAT YOU ALL HAVE FOR TONIGHT'S LECTURE IS NEW YORK TIMES MAGAZINE ARTICLE ON EPIDEMIOLOGY. I INDIRGE YOU TO READ IT. IT'S WRITTEN FOR PEOPLE THAT READ THE NEWSPAPER, IT'S A
17:42
PRETTY NICE COMMON SENSE ARTICLE THAT I THINK EXPLAINS THE BASIS OF WHY WE WANT TO USE EPIDEMIOLOGY VERY WELL WITHOUT OVERSELLING OR UNDER SELLING THE SHORT COMINGS. SO ARE ALL THESE BIG STUDIES REFUTED? THERE WERE A LOT OF EP YES STUDIES PRIOR TO THE WOMEN'S HEALTH INITIATIVE T WOMEN'S HEALTH INITIATIVE WAS A LARGE -- KIND OF A JOKE TO SOME FOLKS THAT WOMEN HAD NOT BEEN STUDIED IN MOST NIH TRIALS MOOCH AND THEN THEY TRIED TO DO ONE TRIAL TO ANSWER ALL QUESTIONS THAT HAD NEVER BEEN ADDRESSED.
18:18
THAT'S NOT EXACTLY WHAT HAPPENED BUT IT ALMOST FELT LIKE THAT. THIS IS A HUGE LOOK AT BOTH SOME EPIDEMIOLOGIC ASPECTS AND RANDOMIZED TRIAL ASPECTS FREEWAY WE DID, IF WE DID HORMONE REPLACEMENT THERAPY WHAT WERE THE IMPACTS ON BONES, CANCER, CARDIAC HEALTH. THERE HAVE BEEN A LOT OF EPIDEMIOLOGIC STUDIES. PEOPLE SAID WE ALREADY KNEW THE ANSWER BUT REALISTICALLY, THOSE STUDIES DID NOT AGREE. AND AFTER THE WOMEN'S HEALTH INITIATIVE WAS STOPPED EARLY ENTHEIR RANDOMIZED TRIALS
18:56
BECAUSE IT WAS SEEN IN EFFECT THERE WERE NOT MANY POSITIVE BENEFITS FOR MOST OF THE OUTCOMES -- THERE WERE FOR SOME, BUT NOT FOR ALL. BY THEN, WOE HAD STRONGER COMPUTING POWER AND FEW STATISTICAL METHODS. WITH THE KNOWLEDGE WE HAD, ROSS WENT BACK AND RENALESED ALMOST ALL -- REANALYZED ALMOST ALL THE DATA USED AND SAID ACTUALLY, WE CAN REPRODUCE THE RANDOMIZED STUDIES. THAT HELPED GIVE CREDENCE TO THAT BUT THE NUMBER ONE THING THAT WAS MISSING WAS THE SOCIOECONOMIC STATUS.
19:31
SO THAT'S THE PROBLEM, AGAIN, WHEN YOU'RE DOING NON RANDOMIZED TRIALS. YOU DON'T ALWAYS KNOW ALL THE VARIABLES YOU SHOULD BE PUTTING IN THE MODEL. NOW, PREVENTION IS SOMETHING THAT WE WOULD LOVE TO STUDY WITH RANDOMIZED STUDIES BUT IT'S VERY HARD FOR ME DO TELL PEOPLE TO ACTUALLY FOLLOW A CERTAIN LIFE STYLE FOR 20, 50 YEARS AND WAIT FOR THE OUTCOME. PREVENTION IS A HUGE PLACE WHERE WE USE EPIDEMIOLOGY. SO THIS IS THE LENGTH TO NEW YORK TIMES MAGAZINE WHERE YOU HAVE A PDF OF THE ARTICLE IF YOU
20:05
WANT TO LOOK AT IT. EPI IS HARD. YOU HAVE TO MEASURE A TON OF THINGS IN A LOT OF DIFFERENT WAYS. AND YOU HAVE TO DO IT VERY ACCURATELY, FAIRLY OFTEN. BASE A LOT OF THESE VARIABLES ARE CHANGING OVERTIME. YOU WANT TO NOT LOSE ANY DATA SO YOU'VE GOT TO TRACK EVERYBODY DOWN AT EVERY TIME POINT. YOU HAVE TO MEASURE IT THE SAME WAY EVERY TIME. THIS IS A PROBLEM BECAUSE OUR METHODS OF MEASUREMENT MAY BE CHANGING. ALSO, WHERE WE MAKE DIAGNOSES MAY BE CHANGING, SO YOU HAVE TO
20:37
WORK ALL THIS INTO YOUR PROTOCOL AND ANALYSES. BUT YOU'RE NEVER GOING TO KNOW WHAT YOU DON'T MEASURE. I DO LIKE THE PHRASE THAT I FORGET WHO USED THIS, THAT EPIDEMIOLOGY IS LIKE HYPOTHESISIS GENERATING EVIDENCE. KIND OF LIKE CIRCLE EVIDENCE IN COURT. CIRCUMSTANTIAL EVIDENCE IN COURT. OR LOOK LIKE WATCHING THOSE TV SHOWS, YOU'RE LIKE I KNOW WHO DID IT. IT WAS REALLY CLEAR BUT THEY HAVE TO PROVE IT METHODICALLY. EPI IS THAT CIRCUMSTANTIAL EVIDENCE. IT ALSO MIGHT BE YOUR ONLY
21:16
EVIDENCE OUTSIDE A LABORATORY. THERE ARE TIMES THAT HE CAN RANDOMIZE PATIENTS BUT I CAN'T RANDOMIZE YOU TO SMOKE OR NOT SPOKE. I CAN RANDOMIZE AN ANIMAL IN THE LAB TO BE EXPOSED TO CIGARETTE SMOKE, YES OR NO. BUT I CAN'T RANDOMIZE HUMANS. SO THERE ARE STILL, AS WE TALKED ABOUT LAST WEEK, FUNDAMENTAL LIMITATIONS ON THESE OBSERVATIONAL STUDIES. I CAN DISTINGUISH AN ASSOCIATION BUT I CAN'T INHERENTLY DETERMINE CAUSALITY, UNLESS I FIGURE OUT ANOTHER WAY. HOW ELSE WHERE I GENERATE HIPSIS?
21:53
I LOVE EPIDEMIOLOGY. IT'S HUGELY CHALLENGING BUT IT'S A REAL CORNERSTONE AND IMPORTANT. IT'S ESPECIALLY IMPORTANT FOR GENERATING HYPOTHESIS, BECAUSE YOU'VE GOT EPI, CLINICIAN EXPERIENCE AND OBSERVATION, WHICH REELISTICLY WHEN WELL DONE IS EPIDEMIOLOGY OR THEY'LL PULLING IT OUT OF THIN AIR. I'LL TAKE EITHER OF THE FIRST TWO, THANK YOU VERY MUCH. I'M GOING TO MENTION A LOT ABOUT CANCER AND SMOKING BECAUSE THERE IS A LOT OF INFORMATION THERE. BUT IT WAS IN THE DISCUSSION OF
22:27
IF SMOKING WAS TIED TO LUNG CANCER, THAT WOE CAME UP WITH A CRITERIA HOW YOU DO CASUAL INFERENCE FROM OBSERVATIONAL STUDIES. THIS GROUP OF FOLKS SET UP A LIST OF EPIDEMIOLOGIC CRITERIA FOR THAT. YOU HAVE TO HAVE STATISTICAL SIGNIFICANCE. IT'S REALLY EASY TO GET STATISTICAL SIGNIFICANCE. PAUL WILL TALK ABOUT THAT IN THE HYPOTHESIS TESTING LECTURE. AND I SAY THAT AND EVERY ONE WHO HAS NEVER GOTTEN IT WAS GOING TO COMPLAIN. IF YOU HAVE A HIGH ENOUGH SAMPLE SIZE YOU'VE FIND STATISTICAL
23:04
SIGNIFICANCE. MAY NOT BE CLINICAL LEE MEANINGFUL BUT YOU'LL HAVE STATISTICAL SIGNIFICANCE. THAT GETS ME TO STRENGTH OF ASSOCIATION. YOU WANT TO HAVE ASSISTANT STRENGTH OF ASSOCIATION THAT'S MEANINGFUL. WHETHER THAT'S AN ODD RATIO, RELATIVE RISK, DOESN'T PARTICULARLY MATTER. YOU WANT TO SEE A DOSE RESPONSE RELATIONSHIP. ONE THING DO CONSIDER IT MIGHT BE A YOU OR UPSIDE-DOWN U. BUT TYPICALLY, AS I INCREASE YOUR EXPOSURE, YOU SHOULD BE HAVING INCREASED RESPONSES. THIS SHOULD BE A TEMP PORIAL
23:39
SEQUENCE. CAUSE HAS TO HAPPEN BEFORE THE EFFECT. YOU WOULD BE SHOCKED HOW MANY TIMES THAT'S WHERE IT FALLS DOWN FOR THE CAUSAL INFERENCE. YOU HAVE TO HAVE THIS CONSISTENCY OF THE ASSOCIATION AND REPLICATION OF THE RESULTS. AND DR. STONEY ARE TALK MORE ABOUT THIS IN HER LECTURE. YOU NEED BIOLOGICAL PLAUSIBILITY. TRUE STORY. I SENT AN EXPLANATION OF SOME NUMBERS DO A PI. AND I HAD RUN THE RUNNING MODEM IT TURNS OUT. -- MODEL. SO HE'S LIKE THIS IS GREAT. HE SENDS ME -- WHEN I FIRST SET
24:19
THE NUMBERS, HE SENDS THIS WHOLE REASON WHY THIS MAKES SENSE. I REALIZE I MADE A MISTAKE. ACTUALLY, IT'S THE INVERSE OF THE NUMBERS THAT I SENT YOU. THAT'S FINE. GIVES ME A WHOLE ANOTHER EXPLANATION FOR IT. WE DON'T ALWAYS KNOW -- HE WAS A TOTALLY GOOD INVESTIGATOR. HE JUST -- THE THING IS, WE DON'T KNOW ENOUGH ABOUT THE BIOLOGICAL. A LOT OF TIMES WE CAN MAKE A VERY PLAUSIBLE STORY GOING IN ALMOST EVERY DIRECTION. SO WHAT I SHOULD BE BIOLOGICALLY PLAUSIBLE, MAYBE WE DON'T KNOW ENOUGH.
24:53
OR WE CAN STILL MAKE THE STORY WORK. BUT THE LAST ONE, THIS H, IS A KEY. EXEXPERIMENTAL EVIDENCE. WHAT DOES THAT MEAN? MAYBE YOU'RE RANDOMIZING PIGS IN THE LAB. MAYBE YOU'RE RANDOMIZING RATS OR MICE. SOMEWHERE, SOMEHOW, YOU DID RUN THE EXPERIMENT. IT JUST MAY NOT HAVE BEEN IN HUMAN BEINGS OR IN THE ANIMAL OR THE PLANT OR WHATEVER THAT YOU WANT TO ACTUALLY RUN THE EXPERIMENT IN. MAYBE IT'S NOT IN THE FOREST, IT'S IN THE [INDISCERNIBLE]. BUT SOMEWHERE SOMEHOW THERE HAS
25:29
BEEN EXPERIMENTATION SUCH THAT YOU'RE ABLE TO SHOW THAT THERE IS EVIDENCE OF WHAT'S GOING ON. BEHIND THE EPI. THIS ISN'T NECESSARY LEE GOING TO BE A STRONG INTERVENTIONAL TRIAL BUT YOU'RE GOING TO HOPE IT'S BETTER THAN NOTHING. THIS GETS TO THE EXPERIMENTAL VERSES OBSERVATIONAL ELEMENT. EPIDEMIOLOGISTS USE A LOT OF DIFFERENT DISEASED. TONIGHT, I THINK -- STUDIES. I HAD LABELED MY LECTURE OBSERVATIONAL STUDIES. I NOTICED IT SAID EPI. AS A SHOUT OUT TO MY FRIENDS AT THE CENTER FOR CDC -- WHY AM I
26:15
BLANKING? BUT IT IS BOTH ALSO PREVENTION AT THE END, THERE IS A P THAT'S NOT IN THEIR NAME. BUT THE ISSUES THEY -- THEY DO A LOT OF INTERVENTIONAL STUDIES. THEY TEND TO NOT BE RANDOMIZED. BUT IT DEPENDS. BUT REALLY WHAT THEY'RE KNOWN FOR CLASSILY ARE THE OBSERVATIONAL STUDIES. BUT THIS IS THE ONE SLIDE AS AN ODE TO WHAT THE CDC HAS. THERE ARE EXTENSIVE CLASSES ABOUT EPIDEMIOLOGY IF YOU WANT TO TAKE THEM. REMEMBER THE INTERVENTIONAL EXPERIMENTAL STUDIES, WHEN THE RESEARCHERS INFLUENCES WHAT
26:48
HAPPENS TO THE SUBJECTS. THERE IS A VERY DIFFERENT CONCEPT OF HUMAN SUBJECTS PROTECTIONS THERE THAN WHEN YOU DO AN OBSERVATIONAL STUDY. AND THAT IS TRIIN ALMOST EVERY SINGLE COUNTRY. YOU STILL HAVE HUMAN SUBJECT PROTECTIONS ISSUES IN OBSERVATIONAL STUDIES. IF YOU'RE IN THE UNITED STATES, YOU'RE STILL GOING TO HAVE TO GO TO AN INSTITUTIONAL REVIEW BOARD OR SOME GROUP LIKE THAT AND AGAIN, TRUE IN ALMOST EVERY COUNTRY. BUT OBSERVATIONAL STUDIES ARE DESCRIBESIVE. YOU'RE NOT INTERVENING ON PEOPLE.
27:23
THAT DOESN'T MEAN THAT BY ASKING THEM QUESTIONS OR TAKING SAMPLES, SOMETHING COULDN'T HAPPEN TO THEM. BUT IN THESE OBSERVATIONAL STUDIES, YOU'RE KIND OF LIKE A REPORTER. YOU TRY TO GATHER FORE. WHO WHAT WHERE WHEN. AND THEN ON THE ANALYTICAL SKIED YOU'RE TRYING TO FIGURE OUT HOW AND WHY. SO NOW WE'RE GOING TO GO THROUGH SOME OF THE STUDY DESIGNS. CASE REPORTS AND CASE SERIES ARE FAIRLY SIMILAR. THIS IS WHEN -- AND THIS IS WHAT I WAS TALKING ABOUT, THE PHYSICIANS OF A GUT FEELING.
27:57
IF THEY'RE RECORDING THE EVIDENCE, CASE REPORTS AND SERIES. OBSERVATIONS OF PATIENTS WITH SOME DECLINED CLINICAL CARE CHARACTERISTIC. THEY MAY HAVE A DISEASE, SEEMS TO BE MORE SEVERE OR THEY DON'T SEEM TO BE PART OF THE POPULATION THAT NORMALLY GETS THIS DISEASE. COULD BE YOU SEE A CLUSTER OF SYMPTOMS THAT SEEM ODD. A INDICATION REPORT IS GOING TO HAVE A DESCRIPTION OF DATA TYPICALLY WITHOUT A COMPARISON GROUP. AND AGAIN, THOUGH, YOU NEED A WELL DEFINED POPULATION OF PEOPLE.
28:30
YOU'RE GOING TO HEAR ME SAY THAT ABOUT 20 TIMES TONIGHT. WHY DOES IT NEED TO BE WELL DEFINED? YOU NEED TO HAVE CLEAR DEFINITIONS OF YOUR PHENOMENON. OTHER PEOPLE ARE NOW GOING TO READ THIS REPORTED, READ ABOUT THIS SERIES OF CASES THAT YOU SAW, AND TRY TO SEE IF THEY HAVE THAT SAME GROUP. SO YOU NEED TO BE CLEAR ABOUT WHAT YOU KNOW AND WHAT YOU DON'T KNOW. BECAUSE YOU'RE TRYING TO REALLY DEFINE A DISEASE. WHAT ARE THE DEFINITIONS THAT PUT INDIVIDUALS IN DESIRES AND
29:00
NOT IN DESIRES? THAT'S WHAT YOU HAVE TO DEFINE AND MAKE CLEAR SO OTHER PEOPLE CAN REPLICATE THAT WITH THEIR OWN PATIENTS. THE OBSERVATIONS YOU MAKE NEED TO BE RELIABLE AND REPRODUCIBLE. SOMEBODY ELSE IS GOING TO TRY TO SAY HEY, DO MY PATIENTS HAVE THAT BLOOD LEVEL? DO MY PATIENTS HAVE THIS CHARACTERISTIC? DONE WELL THESE ARE EXTREMELY USEFUL. SO WHAT TYPE OF MATH DO I TEND TO SEE? I SEE MEANS AND MEDIANS. STANDARD DEVIATION OF THE ERROR. A LOT OF TIMES PEOPLE PUBLISH THE DATA.
29:36
YOU'LL SEE THIS TABLE, SUBJECT ONE THROUGH 20. HERE IS THEIR AGE, HERE IS THEIR SECTION, HERE IS ALL THIS -- SEX, HERE IS ALL THIS OTHER STUFF. BECAUSE IF YOU HAVE A SMALL NUMBER -- IF YOU'RE UNDER LIKE SAMPLE SIZE OF 10, STANDARD DEVIATIONS AND ERRORS WHILE YOU CAN CALCULATE THEM, ARE MEANINGLESS. BUT YOU MAY SEE PROPORTIONS. DO YOU ONLY HAVE ONE FEMALE? WELL, THAT'S PRETTY IMPORTANT IF YOU HAVE 20 FOLKS. STUFF LIKE THAT YOU'RE INTERESTED IN. SHOULD HAVE MOVED THE CONFIDENCE
30:09
LIMITS AND INTERVALS BY THE STANDARD DEVIATION. SAME RULE APPLIES, IF YOUR SAMPLE SIZE IS UNDER 10 OR 15, THAT NUMBER IS PRETTY NOT VERY RELIABLE. WE'VE ALWAYS GOT SEPARATE DATA FOR SUBGROUPS SOMETIMES. SOMETIMESTHEY'LL BREAK IT OUT. RURAL VERSES URBAN, MEN VERSES WOMEN, YOUNGER VERSES OLDER. THEY SEEM TO HAVE THE SAME PHENOMENON. THINK ABOUT A REALLY BAD INFLUENZA. IT MIGHT IMPACT THE VERY YOUNG AND OLD. I MAY SAY I'VE GOT THESE 2 DIFFERENT GROUPS, I MAY PUT THEM
30:42
NEXT TO EACH OTHER ON THE TABLE. BUT WHAT MAY BE REALLY THE ODD CHARACTERISTIC IS SUDDENLY, THE ONLY PEOPLE DYING FROM FRIENDSA ARE ENTHEIR 30s AND WERE RELATIVELY HEALTHY TO BEGIN THAT. THAT'S THE TYPE OF THING YOU PUT INTO A CASE SERIES, AND ALSO CHARACTERISTICS THAT HAVE BEEN DEFINING YOUR GROUP. SO WHAT DO WE USE THESE FOR SNOW YOU'VE GOT HYPOTHESES FORMATION, WE HAVE NATURAL HISTORY SO THE BEGINNING OF NATURAL HISTORY STUDIES ARE MANY TIMES THESE CASE SERIES OR CASE REPORTS.
31:13
THIS IS BASICALLY DEFINING YOUR CLINICAL EXPERIENCE TAKING CARE OF THESE PATIENTS. DO YOU HAVE BIASED PATIENT SELECTION? YES. ALMOST EVERY CASE SERIES THAT'S PUBLISHED, IT'S A SET OF PATIENTS THAT LIKE ONE GROUP SAW. IT MIGHT BE JUST A BLIP. BUT THE IDEA IS TO FILLING OUT THE GENERALIZABILITY OF THESE RESULTS. IF IT IS SOMETHING THAT IS GENERALIZABLE TO LARGER SETS OF CLIPPINGS AROUND THE WORLD, WE MAY HAVE AN ACTUAL EPIDEMIC ON OUR HANDS. IT COULD BE CHANCE, IT COULD BE A CHARACTERISTIC.
31:49
AT THIS POINT WE DON'T KNOW. WHICH IS WHY WE'RE TRYING TO RECORD EVERYTHING AND FIGURE IT OUT. IT'S ONLY WITH CAREFUL MEASUREMENT AND RECORDING THAT WE CAN ACTUALLY GET THIS WORK DONE WELL. AN EXAMPLE OF WHERE IT WAS DONE VERY WELL, AGAIN, FROM 1981, WAS THE INITIAL REPORT OF FIVE CASES OF PNEUMO CYSTIC PNEUMONIA IS PREVIOUSLY HEALTHY HOMO SEXUAL MEN. THE FIRST TIME THE CENTERS FOR CITIES CONTROL AND PREVENTION PUT SEXUAL ORIENTATION IN THE DISCUSSION. LET ALONE IN THE TITLE OF
32:24
SOMETHING IN MORBIDITY MORTALITY WEEKLY'S REPORT. ACTUALLY, THIS WAS NOTICED HERE AT THE NIH ALSO. BUT WHERE THIS REPORT STARTED FROM WAS A PHARMACIST AT THE TIME THIS TIME OF PNEUMONIA WAS TYPICALLY SEEN IN ADVANCED CANCER PATIENTS. IN ORDER TO GET THE DRUG IN THE UNITED STATES IT HAD TO GO THROUGH A CENTRAL PHARMACY. THAT LADY LITERALLY, ONE PERSON IN CHARGE OF FILLING ALL OF THESE, STARTED NOTICING THAT THE PATIENTS THAT SHE WAS GETTING THE REQUESTS FOR THIS DRUG FOR
32:56
WERE VERY DIFFERENT THAN THE PATIENTS THAT SHE HAD ALWAYS GIVEN THE DRUG TO. AND THEY STARTED AN INVESTIGATION. AND WHEN THEY DID, THEY -- YES, THE PATIENTS HAVE THIS TYPE OF PNEUMONIA PNEUMONIA, THEY NEEDED TO BE TREATED. SOMETHING ELSE IS IMPAIRING THEIR IMMUNE SYSTEM. IT WAS NOT THEIR CANCER TREATMENTS. AND THAT'S WHAT LED TO THIS REPORT. AND THEN ADDITIONAL REPORTS THAT STARTED COMING IN FROM ACROSS THE UNITED STATES AND ACROSS THE WORLD THAT PEOPLE WERE ALSO SEEING FOLKS THAT MATCHED THESE
33:27
CHARACTERISTICS AND THEY EVENTUALLY CAME UP WITH A BETTER DEFINITION. SO, NEVER UNDERESTIMATE IN YOUR CLINICAL PRACTICE, IN THAT CLINICAL PRACTICE INVOLVES ALL TYPES OF CLINICAL WORK. WHAT YOU MIGHT SEE AND HOW IMPACTFUL THAT CAN BE. WHETHER IT'S H1N1, THE NEXT EBOLA OUTBREAK OR WHATEVER IS GOING TO COME. SOMETHING WILL COME. THIS WORK IS IMPORTANT. SO NOW WE'RE GOING TO GET TO CROSS SECTIONAL SURVEYS. HERE IS WHERE WE'RE GOING TO OBSERVE THE PREVALENCE OF CHARACTERISTICS OF DISEASE.
34:08
CROSS SECTIONAL REMEMBER IS THE SNAPSHOT IN TIME. AGAIN, MY IDEA IS I NEED TO UNDERSTAND THE CHARACTERISTICS OF PARTICIPANTS IN MY DECIDE IN A WELL DEFINED POPULATION. FIRST THINGS FIRST. DEFINE YOUR POPULATION. AM I GOING TO LOOK AT ALL ADULTS 18 AND OLDER IN THE UNITED STATES? AM I GOING TO TRY TO FIND EVERYBODY THAT HAS A VERY SPECIFIC RARE DISEASE AROUND THE WORLD? AM I GOING TO TRY TO FIND A SAMPLE OF PEOPLE WITH THIS DISEASE ON THE AFRICAN CONTINENT? DEFINE YOUR POPULATION.
34:54
DERIVE A SAMPLE OF THAT POPULATION. WHO ARE YOU GOING TO ASK AND GIVE THE SURVEY TO? IT'S PROBABLY NOT EVERYBODY. BUT THEN YOU HAVE TO MAKE SURE. DO I HAVE REPRESENTATIVENESS IN WHO I'M GIVING THE SURVEY TO? IF I ONLY GIVE THE SURVEY TO MEN I CAN'T REALLY TALK ABOUT WOMEN. IF I ONLY GIVE IT TO THREE WOMEN, I PROBABLY STILL CAN'T TALK ABOUT WOMEN. HAVE I COVERED THE AGE RANGES, THE DIFFERENT CULTURAL ISSUES THAT COULD BE COMING UP? AND A LOT OF OTHER ISSUES. IF YOU'RE GETTING NIH FUNDING,
35:29
YOU HAVE TO REPORT ON YOUR SAMPLING ABOUT RACE, ETHNICITY, AND SEX. YOU'LL GET A WHOLE LECTURE ON THAT. YOU HAVE TO DEFINE THE CHARACTERISTICS THAT ARE BEING STUDIED. WHAT ARE THE OBSERVATIONS YOU'RE MAKING, WHAT ARE THE QUESTIONS AND RESPONSES TO THE SURVEY AND WHAT ARE THE INSTRUCTIONS. >> YOU SAY YOU'RE MEASURING BLOOD PRESSURE. WAITED HOW? ARE YOU TAKING A SPECIFIC SCALE TO PEOPLE'S HOUSES, MAKING SURE THAT IT'S BEEN PROPERRY ADJUSTED AND WILL MEASURE CORRECTLY?
36:03
ARE YOU GOING TO DO IT ON THE FLOOR, ON CARPET? HOW IS THAT GOING TO CHANGE YOUR RESULT? ARE YOU JUST GOING TO ASK PEOPLE HOW MUCH THEY WEIGH? ARE YOU GOING TO DO IT IN KILOGRAMS IS THIS OR IN POUNDS? HOW MANY DECIBEL PLACES ARE YOU GOING OUT, ARE YOU NOT? YOU MUST HAVE STANDARDIZED OBSERVATIONS, YOU MUST TRAIN THE PEOPLE MAKING THE OBSERVATIONS ON HOW TO DO IT. AND THAT'S YOUR WAY TO REDUCE THE MEASUREMENT ERROR. IF YOU'RE ASKING FOR SELF-REPORT THAT'S OKAY, BUT YOU HAVE TO
36:32
AMTHAT WHEN YOU'RE DOING -- KNOWLEDGE THAT WHEN YOU'RE DOING YOUR DECIDE DESIGN. YOU NEED TO HAVE CLEARLY DEFINED CHARACTERISTICS. REMEMBER SLEEP. I NEED TO KNOW MORE. NOT JUST HOW DID THEY SLEEP, YOU NEED TO HAVE VERY DEFINED QUESTIONS. WHEN DO I GO TO BED? HOW LONG HAVE YOU SLEPT. HOW WAS THAT MEASURED? THOSE METHODS MUCH TO BE APPLIED EQUALLY TO YOUR STUDY PARTICIPANTS. IF YOU HAVE A QUESTION YOU CAN ASK PEOPLE WHO ARE SICK BUT NOT ASK PEOPLE WHO ARE WELL, THAT'S A PROBLEM.
37:08
WHAT I AM -- WHEN I'M TRYING TO DO A SURVEY I NEED TO FIGURE OUT HOW TO GET THE INFORMATION FROM EVERYBODY. SOMEBODY ILLITERATE WHO CAN'T READ AND I HAND THEM A SHEET OF PAPER TO FILL OUT THAT'S A PROBLEM. SO IF I I CAN'T DO THIS SAME METHOD OF DATA COLLECTION, IODAMIDE TO, THEN, BE ABLE TO KNOW THAT MY MULTIPLE METHODS OF DATA COLLECTION WILL NOT BIAS THE RESULTS. THAT'S A BIG ISSUE. WHAT ABOUT PREVALENCE AND INCIDENTS? I MENTIONED PREVALENCE. PREVALENCE IS THE NUMBER OF
37:48
PEOPLE WITH DISEASE DIVIDED BY THE NUMBER OF PEOPLE AT RISK. LIKE TAKING A SNAP SHOP. HOW MANY DIABETES, TYPE TWO FROM BRAZIL RIGHT NOW, THAT'S PREVALENCE. PREVALENCE IS THE INCIDENCE TIMES DURATION. SO WHAT IS INCIDENCE. >> INCIDENCE IS YOUR NUMBER OF NEW CASES OF DISEASE OVER SOME PERIOD OF TIME. DIVIDED BY THE NUMBER OF PEOPLE WHO WERE AT RISK AT A PERIOD. IF I ALREADY HAVE TIME TO YOU DIABETES, I'M NOT AT RISK OF GETTING IT AGAIN. SO AN EXAMPLE, HOW MANY NEW INCIDENT CASES OF TYPE TWO
38:27
DIABETES WERE DIAGNOSED IN 2007 DIVIDED BY THE PEOPLE THAT CAN DEVELOP THE DISEASE IN 2007. SO WHAT'S GOING TO BE DESCRIBED IN TABLES? YOU'RE LOOKING AT THESE CROSS SECTIONAL SURVEYS YOU GET A LOT OF DESCRIPTIVE QUESTIONS. YOU WANT TO BASICALLY TELL FOLKS HOW COMMON ARE EACH OF THESE DIFFERENT CHARACTERISTICS OR FACTORS. >> WHAT'S THE DISTRIBUTION OF THE FACTORS OF INTEREST? THINK ABOUT AGE. LIKELY TO GET THIS YOUNG, OLD, SOMEWHERE IN BETWEEN? THAT'S WHAT GETS DESCRIBED IN
39:01
THE TABLES IN -- DISCUSSING THESE SURVEYS. YOU ALSO ARE LOOKING FOR THESE ASSOCIATIONS. WHAT'S THE RELATIONSHIP BETWEEN THE FACTORS? AND HOW DO PEOPLE THAT HAVE ONE FACTOR DIFFER FROM THOSE WITHOUT? SO THIS GETS INTO THAT IDEA, AGAIN, OF CONFOUNDING AND EFFECT MODIFICATION. SO I MIGHT HAVE SOME PERCENT -- THEY MY TALK ABOUT PREVALENCE. SOMETIMES THEY GIVE YOU A NUMBER LIKE 65 CASES PER 10,000 LIFE YEARS OR SOMETHING LIKE THAT, LIKE YOU'LL SEE DIFFERENT NUMBERS. OR 10,000 PEOPLE.
39:43
THE LIFE YEARS TENDS TO BE MORE OF THE INCIDENCE. MEAN OR MEDIUM LEVELS OF RELATIVE FACTORS. SUBSET THE DESCRIPTIVE INFORMATION BY IMPORTANT SUBGROUPS. IF IT'S HIV YOU MAY DO SOMETHING DIFFERENT FOR INTRAVENOUS DRUG USERS, NON DRUG USERS, PEOPLE HAVE DIFFERENT SEXUAL BEHAVIORS, YOU MIGHT DO DIFFERENT TABLES. FOR THE ANALYSES, YOU TEND TO SEE A LOT OF CATEGORICAL ANALYSES. YOU WANT TO USE REGRESSION. THANKFULLY, THE JOURNALS ARE GETTING PICKIER ABOUT THIS. ALL THESE SQUARE TESTS.
40:22
BASICALLY LIKE OH, WHAT'S THE RELATIONSHIP BETWEENTH BODY MASSES INDEX CATEGORY AND THEIR SEX. THEY WOULD COMPARE THESE TWO VARIABLES AT EACH GIVEN TIME INSTEAD OF WRITING A REGRESSION. SO I WANT FOLKS TO USE LOGISTICAL OR ORDINAL REGRESSION, MANY TIMES, FOR THEIR COVARIABLES SO IT CAN ADD IN ALL THE POTENTIAL CONFOUNDERS AND EFFECT MODIFIERS, AND SEE THE BIG PICTURE FOR FOLKS. IF YOU HAVE A VARIABLE YOU'RE LOOKING AT YOU MIGHT DO T. TESTS BUT AGAIN YOU SHOULD PROBABLY BE
40:59
USING A FORM OF REGRESSION INSTEAD. SO REMEMBER IF ALL ELSE FAILS REGRESSION IS YOUR FRIEND. SO I FORM OF REGRESSION. BUT SINCE EVERYBODY USES 2 BY 2 TABLES I FIGURE I SHOULD TEACH IT TO YOU A LITTLE BIT. YOU'VE GOT THIS REPRESENTATIVE SAMPLE AND YOU CLASSIFY THEM AND WE SAY OUTCOME, BUT WE DON'T KNOW THAT THOSE ARE DIRECT CAUSAL RELATIONSHIP. LET'S SAY PEOPLE HAVE A DISEASE OR DIDN'T HAVE A DISEASE. AND THEN THEY DID OR DIDN'T HAVE A RISK FACTOR. THINK BACK TO JOHN GALLIN'S TALK.
41:31
I THINK HE TALKED ABOUT THE CLASSIC EXAMPLE WITH THE WELLS. AND YOU COULD SAY DID YOU GET YOUR WATER FROM THIS WELL? YES OR NO. DID I GET THIS LOVELY DISEASE, YES OR NO. CLASSIFY PEOPLE IN THESE 4 WAYS. THERE WAS A STUDY OUT OF AFRICA LOOKING AT THE EFFICIENCY OF THE BCG VACCINATION AND PREVENTING TB. THE NICE THING IS THAT I RESPECT LEAVES A MARK ON YOUR ARM OR THE OLD VERSION DID. THE NEW VERSION DOESN'T SO YOU CRONET DO IT NOW. SO THEY BASICALLY LOOKED, SAID OKAY WE'RE GOING TO TAKE A
42:09
SAMPLE OF PEOPLE, DID THEY HAVE THE VACCINATION, DO THEY HAVE THE MARK FROM THE VACCINE INTERSECTION, YES OR NO? AND IF NOT, THEN THAT -- IF THEY DIDN'T -- I'M SORRY. THEN THEY LOOKED TO SEE IF THEY HAD TB, YES OR NO. SO WHAT'S YOUR POSITIVE ATTRIBUTE? THE NICE THING ABOUT THESE SURVEYS, THEY'RE CHEAP. IF YOU HAVE A COMMON DISEASE. SO IF LOTS OF PEOPLE WILL HAVE THE DISEASE I COULD PROBABLY FIGURE OUT A LOT OF INFORMATION ABOUT IT. LIKE A COMMON INFLUENZA, COMMON COLDS, THINGS LIKE THAT. YOU
42:46
CAN GET A LOT OF INFORMATION VERY INEXPENSIVELY. IT'S MORE REPRESENTATIVE OF CASES THAN THAT CASE SERIES WE WERE TALKINGS ABOUT. WHY? I'M TALKING A BROADER POPULATION. A LOT OF THESE LAST LESS THAN A YEAR. YOU HAVE A VERY SPECIFIC POPULATION THAT YOU'RE LOOKING AT. AND YOU HAVE SIMULTANEOUS WIDE VARIETY OF MEASUREMENTS. THE NEGATIVES. IF YOU HAVE A RARE DISEASE, YOU MIGHT NOT PICK IT UP. SO I INTERVIEW 50,000 PEOPLE. IF ONE IN A MILLION HAVE THIS DISEASE, I PROBABLY AM NOT GOING
43:22
TO HAVE 4 OR 5 PEOPLE IN MY STUDY. IF IT'S UNSUITABLE FOR A DISEASE OF SHORT DURATION, SO I MENTIONED THE COMMON COLD. HOW LONG DOES THAT COLD LAST? MAYBE FIVE DAYS OR SOMETHING? ARE YOU GOING FEW REMEMBER IT WHEN I -- DO REMEMBER IT WHEN I ASK IF YOU HAD A COLD IN THE LAST YEAR AND IT'S BEEN 8 MONTHS? PROBABLY NOT. PROBABLY NOT GOING TO REMEMBER THAT MUCH ABOUT HOW YOU FELT. YOU TEND TO GET HE HIGH REFUSABLE RATES. DO YOU WANT TO FILL OUT THAT 8 PAGE SURVEY? IN A LOT OF CAN YOU BE COUNTRIES
44:02
THIS TYPE OF SURVEY WORK THEY DO, THEY MAKE IT MANDATORY. THEY'LL THROW YOU IN JAIL IF I DON'T ANSWER THE SURVEY. NOT ALL PLACES. TIME IS OUR BEST AND WORST CONFOUNDER. SO TRYING TO REMEMBER INFORMATION CAN BE QUESTIONABLE. SEE EXAMPLES WHERE THIS WAS PRETTY GOOD, WHERE THERE WAS INTERESTING MATH OF THE WE HAD THIS STUDY ON THE PREVALENCE OF OVERWEIGHT AND OBESITY IN U.S. CHILDREN iADOLESCENCE AND ADULTS. THIS WAS PUBLISHED OUT OF THE CENTER OF DISEASE CONTROL AND PREVENTION, NATIONAL HEALTH AND
44:55
NUTRITION INFORMATION. NHANES DATA. THIS IS LAUNCHED EVERY YEAR BY THE CBS. THEY MAKE YOU GET ON A SCALE AND TAKE YOUR WEIGHT AND BLOOD PRESSURE. SO ANOTHER STUDY CAME OUT OF THE CDC AND NATIONAL CANCER INSTITUTE. EXCESS DEATHS ASSOCIATED WITH UNDER WEIGHT, OVER WEIGHT AND OBSAT THAT. SO MOST PEOPLE WILL THINK IF YOU'RE OVERWEIGHT OR OBESE, THAT'S TIED TO HIGHER MORE MORTALITY. PROBLEM IS, THE ANALYSIS DONE LEFT OUT THE MOST IMPORTANT THING, THAT'S YOUR AGE. WHEN I LOOK AT MORTALITY,
45:40
TYPICALLY YOU WANT AGE AND AGE SQUARED IN THE MODEL. BACK IN THE 80s A LOT OF WORK HAD BEEN DONE. THEY SAID BMI AROUND 28 IS PRETTY GOOD, IF YOU'RE THINKING ABOUT MORTALITY. YOU NEED TO TAKE INTO ACCOUNT SMOKING AND ALL THESE FACTORS. BUT THEN THIS HUGE LANDMARK CASE, I WILL NOT LIE, GOT OUR DEPARTMENT A LOT OF MONEY. BUT THESE FOLKS WERE GOING THROUGH THE ANALYSIS AND THE MODEL AND THEY'RE LIKE WHERE IS AGE? WELL, WE DON'T HAVE AGE IN THE MODEL. WELL, AGE IS THE BEST PREDICTOR OF MORTALITY.
46:16
YOU'VE GOT TO PUT IT IN YOUR MODEL. ONCE THEY PUT AGE IN THE MODEL, YOU GOT VERY DIFFERENT RESULTS. FOR SPECIFIC EXCESS DEATH. KATHY HERE, ALONG WITH OTHER AUTHORS, DID A SERIES OF PAPERS, IF YOU COME DOWN TIGHT AND IT WENT THROUGH EXTREME SCRUTINY FROM THE CDC AND NCI BEFORE IT GOT PUBLISHED, BMI AROUND 28 IS BETTER. SO IF YOU HAVE NORMAL WEIGHT -- RATHER, I SHOULD SAY OVERWEIGHT CAN BE PROTECTIVE. AND THERE ARE A ROT OF REASONS PEOPLE THOUGHT THAT MIGHT BE. BUT ACTUALLY, MOST OF THEM WERE
46:58
REFUTED BY THAT DATA. SO IT'S A VERY INTERESTING READ. IF YOU WANT TO LOOK AT THAT AND ALL THE COMMENTARY THAT CAME AFTERWARDS. SO DON'T FORGET TO PIT VARIABLES IN YOUR MODEL. AGE, SEX, THESE ARE EASY PECKINGS, FOLKS, THOSE SHOULD BE PRETTY MUCH EVERY MODEL. NOW, CASE CONTROL STUDIES. ACTUALLY, LET ME BACK UP FROM WHAT I JUST SAID WHICH IS SOMETIMES, IT DOESN'T MATTER. SO EVERYBODY THINKS SEX MATTERS AND YOU LOOK AT IT, PUT IT IN THE MODEL, TAKE IT OUT, AND NOTHING CHANGES. OR AGE, AGE SQUARED.
47:32
USUALLY, THESE ARE VARIABLES YOU SHOULD COLLECT AND ALWAYS EVALUATE IN YOUR MODELS, BUT IN FACT IT MAY NOT MATTER. SO YOU HAVE TO LET THE DATA TELL YOU THE STORY. ONE OR MORE RISK FACTORS AND EXPOSURES. WHEN YOU HAVE A RARE DISEASE, THIS IS WHERE YOU START. HERE, NATURAL HISTORY STUDIES THAT TYPICALLY YOU'RE DOING A CASE CONTROL STUDY. IF YOU HAVE DISEASES WITH LONG LATENCY PERIODS, YOU ALSO TEND TO SEE CASE CONTROL DISEASED. THERQUICKER AND LESS EXPENSIVE THAN THE CORE HOARD CITIES THAT
48:31
WE'RE GOING TO TALK ABOUT AFTER. THIS IS THE PREVALENCE OR LEVEL OF A POSSIBLE RISK FACTOR BETWEEN REPRESENTATIVE GROUP OF DISEASE SUBJECTS, SO YOUR CASES, AND DISEASE FREE SUBJECTS, YOUR CONTROLS. THEY'RE GOING TO HOPEFULLY COME FROM THE GENERAL SAME POPULATION. WHY IS THAT? WELL, BECAUSE I'M TRYING TO CONTROL FOR THE SOCIOECONOMIC FACTORS, ALL THIS OTHER STUFF THAT COULD BE GOING ON. BUT AGAIN WHAT YOU'RE NEVER REALLY SURE OF IS WHAT'S THE RISK FACTOR. YOU HAVE TO BE CAREFUL WHAT
49:06
YOU'RE CONTROLLING IN THAT YOU'RE NOT OVERNIGHT CONTROLLING IN STUDIES. AND A LOT OF CASE CONTROL STUDIES WILL SEE MULTIPLE GROUP CONTROLS TO TRY TO TEASE THIS OUT. BUT THE WORD REPRESENTATIVE IS REALLY KEY HERE. NEED TO MAKE SURE THAT YOUR CASES AND CONTROLS IN PARTICULAR, YOUR CASES, REPRESENT THE DISEASE UNDER STUDY. THIS CAN BE HARD. REALISTICALLY, THE CASES ARE ALL OVER THE PLACE. YOU HAVE TO MAKE SURE THAT THE FACT YOURS YOU'RE TEASING OUT ARE NOT JUST DUE TO LOCATION.
49:48
THIS THIS IS A SOMEPICTURE. YOU'RE GOING TO DEFINE EVERYBODY BY THEIR DISEASE STATS. THAT MEANS YOU HAVE TO BE ABLE TO DEFINE DISEASE AND NO DISEASE. YOU HAVE TO MAKE SURE YOUR NO DISEASE FOLKS HAVE GO DISEASE. THEY MAY NOT KNOW THAT THEY HAVE A DISEASE. YOU'VE GOT TO BE CAREFUL ABOUT THAT. THEN WE HAVE TO FIGURE OUT WHO HAD EXPOSURES AND DIDN'T. A LOT OF TYPES IT WILL LOOK EXACTLY THE SAME ALL THE WAY ACROSS. AGAIN, THESE CASES REPRESENT ALL CASES OF DISEASE. CONTROLS SHOULD REPRESENT A
50:33
HEALTHY POPULATION. THEY MAY NOT BE THAT HEALTHY, BUT NON DISEASED, AGAIN POPULATION. AND THE INFORMATION HAS TO BE COLLECTED IN THE SAME WAY. IF I HAVE PEOPLE WHO HAVE SERVE ISICAL DYSPLASIA, AND I'M GOING TO TAKE A BIOPSY OF THROUGH, THE PEOPLE THAT DON'T HAVE DYSPLASIA, AM I ALLOWED TO TAKE A BIOPSY? WHERE DO I TAKE IT? YOU HAVE TO THINK ABOUT HOW YOU'RE COLLECTING THIS INFORMATION. IT'S ALSO HARD BECAUSE SOMETIMES YOU HAVE INTERVIEWERS, AND I THINK BARBARA WILL TALK ABOUT THIS SOME, TOO.
51:09
YOU'LL HAVE THESE INTERVIEWERS, AND THEY HAVE AN IDEA, LIKE, OKAY, THE CASES TEND TO HAVE HEAD ACADEMICS, SO IF A CASE DOESN'T REPORT A HEAD IKE THEY'LL DIG IN A LITTLE BIT MORE. YOU DON'T WANT THAT TO HAPPEN. YOU WANT TO HAVE THE SAME INFORMATION COLLECTION PROCEDURES. SOME STUDIES WILL AUDIO OR VIDEOTAPE ANY INTERVIEWS THAT HAPPEN TO TRY TO MAKE SURE THAT THIS IS GOING TO BE CONDUCTED WELL, AND THIS THEY'RE GOING TO DO CHECKS ON THEIR STUDY PERSONNEL TO MAKE SURE THAT DATA
51:40
COLLECTION IS CONSISTENT ACROSS THEIR TYPES OF PATIENTS. I ALSO CAN HAVE A VARIETY OF ISSUES THAT COME UP THINKING ABOUT THE REPORTING, WE'LL TALK ABOUT BIAS AT THE END OF TODAY'S LECTURE. BUT IF YOU'RE CHILD WAS BORN AND HAS A DEFECT OF SOME TYPE, OR MAYBE THE CHILD IS BORN AND STARTS TO HAVE REALLY HORRIBLE ILLNESS PROBLEMS, SECOND OR THIRD YEAR OF LIFE. YOUR PROBABLY SITTING UP ALL NIGHT THINKING ABOUT WHAT DID I DO DURING MY PREGNANCY? WHAT DID I EAT? WHAT DID MY CHILD EAT?
52:14
WHAT WERE THEY EXPOSED TO? YOU'RE THINKING OF ALL OF THIS STUFF BECAUSE YOU'RE NOT SLEEPING, AND YOU'RE READING DIFFERENT THINGS, YOU'RE DRYING TO FIGURE OUT WHAT CAUSED THIS. IF MY CHILD IS PERFECTLY NORMAL, I'M PROBABLY NOT THINKING ABOUT THIS AS MUCH. I'M NOT RECORDING EVERYTHING THAT THEY DO AND EVERYTHING THEY ATE AND EVERYTHING I CAN THINK ABOUT. SO SOMETIMES THAT RECALL BIAS, DIFFERENT [INDISCERNIBLE] THAT HAPPENS, IS SOMETHING THAT ALSO IMPACTS OUR INFORMATION COLLECTION.
52:46
WE HAVE TO FIGURE OUT WAYS TO TRY TO DO THAT COLLECTION BY -- TO MINIMIZE THAT BIAS THAT CAN HAPPEN. SO OUR CASES ARE AGAIN WELL DEFINED POPULATION. STANDARDIZED COLLECTION CRITERIA AS TO WHAT MAKES A CASE. SOMETIMES WE DO NESTED CASE CONTROL CITIES. WE TAKE ABOUT THAT MORE IN THE BOOK. MORE THAN GROUPS ARE USUALLY NESTED WITH YOUR CASES AND CONTROLLED INSIDE A LARGE COHORT STUDIES. WHERE DO I FIND CASES? SOMETIMES WE HAVE CASE REGISTRIES. BUT AGAIN, WHO IS GOING TO GET REGISTERED?
53:23
YOU NEED TO ONGOING ABOUT THE POTENTIAL BIAS AND REPRESENTATIVENESS OF THE REGISTRY THAT YOU'RE USING. WE MAY LOOK AT ADMISSION RECORDS. MAYBE YOU'RE TRYING TO TRACK DOWN SOME HOSPITAL ISSUE. YOU MAY LOOK AT ADMISSION RECORDS. OR PATHEDOLOGY LOGS. THERE ARE PROBLEMS WITH EVERYTHING METHOD. THE TRICK IS TO FIGURE OUT WHICH METHOD IS GOING TO WORK BEST FOR THE QUESTION THAT YOU HAVE. THE OTHER TRICK IS THAT CASES TEND TO PARTICIPATE, WHY? THEY WANT ANSWERS. IT'S PRETTY EASY TO SELL A CASE
53:55
TO BE IN A CASE CONTROL STUDY. THEY HAVE A LOT ON THE LINE. THEY WANT TO KNOW WHAT'S GOING ON. THERE IS NO PERFECT CONTROL GROUP BUT YOU CAN TRY TO GET ONE. A LOT OF TIMES YOU'LL SEE MATCHING ON DIFFERENT VARIABLES THAT THEY THINK AREN'T INVOLVED TO TRY TO GET YOUR CONTROLS AS SIMILAR TO THE CASE AS POSSIBLE. YOU NEED STANDARDIZED COLLECTION CRITERIA. YOU MAY HAVE MULTIPLE CONTROL GROUPS. SOMETHING YOU'LL SEE GEN POP. THAT MEANS GENERAL POPULATION. THEY MAY FULL SPASMS PER THE
54:34
GENERAL POPULATION -- SAMPLES FROM THE GENERAL POPULATION. YOU MAY PULL PEOPLE FROM A NEIGHBORHOOD. IT COULD BE THE NEIGHBORHOOD THEY GREW UP IN OR THEY LIVE IN NOW. THAT'S A WAY TO CONTROL FOR A LOTS OF ENVIRONMENTAL FACTORS THAT YOU MAY NOT BE ABLE TO ASK ABOUT. FAMILIES. SOMETIMES WILL DO A TWIN STUDY. ONE TWIN IS EFFECTIVE WITH THE DISEASE, ANOTHER IS NOT. SO WE'LL TRY TO COMPARE THEM TO EACH OTHER. AND WE MAKE IT HOSPITAL CONTROLS. PEOPLE WHO WERE VERY SICK AT THE
55:09
SAME TIME BUT DID NOT HAVE THE EXACT DISEASE OF INTEREST. IT'S VERY COSTLY TO OBTAIN CONTROLS. ESPECIALLY WHEN YOU HAVE MULTIPLE CONTROL GROUPS. NOT UNCOMMON TO SEE A HOSPITAL AND NEIGHBORHOOD CONTROL, FOR EXAMPLE. WE'RE GOING TO DO ADJUSTMENTS OF RESULTS DURING THE ANALYSIS, AGAIN, YOU WANT TO USE REGRESSION ISSUES NOT YOUR SQUARE TESTS. BUT ALSO REMEMBER CONTROLS -- IT'S ONE THING IF IT'S YOUR TWIN SISTER. YOU MIGHT BE MORE LIKELY TO PARTICIPATE. OTHERWISE, I LIVE A BUSY LIFE.
55:45
DYSPLASIA DO I WANT TO BE A -- WHY DO I WANT TO BE A CONTROL IN YOUR STUDY? IT'S HARD TO GET CONTROLS. A LOT OF TIMES YOU HAVE TO HAVE 8 CONTROLS BEFORE YOU GET WITHIN TO PARTICIPATE. IT'S GENIUS IN MY MIND BUT THAT IS -- GENEROUS IN MY MIND, BUT THAT'S WHAT THE LITERATURE TELLS YOU. LING IO SEED A LOT OF THE SAME METHODS THAT YOU SEE FOR OTHER DISEASED. THE VALTY, THE MEASUREMENT TECHNIQUES ARE IMPORTANT, ALL THAT OTHER STUFF. YOU HAVE THE BIASES THAT CAN HAPPEN. SELECTION OBSERVATION
56:18
INTERVIEWER, ET CETERA. WE TRITO TRY TO SIMILARS ARE IN A 2 BY 2 TABLE. PLEASE DON'T. THIS IS WHAT THE 2 BY 2 TABLES LOOK LIKE. PRESENCE ABSENCE CHARACTERISTICS, OF DISEASE. THE ONLY REASON I'M SHOWING YOU THESE LETTERS BEYOND THE FACT THAT IT'S IN MOST EPIDEMIOLOGY TEXTBOOKS, IS BECAUSE THOSE LETTERS ARE USED TO TRY TO SHOW HOW YOU CALCULATE ODDS RATIOS, RELATIVE RISK, ET CETERA. I WANTED YOU TO HAVE THE PICTURE. THE ANALYSES TEND TO BE PROPORTIONATE CASES EXPOSED COMPARED TO THE PROPORTION OF
56:54
PROPOSALS EXPOSED. A LOT OF TIMES WHILE WE USE ALL THESE DIFFERENT TESTS IT SHOULD COME DOWN TO REGRESSION. WHEN IT DOUBT, NEWS REGRESSION. USE REGRESSION. PUT A BUNCH OF VARIABLES IN THE MODEL. POSITIVE ATTRIBUTES. IF YOU WANT TO STUDY RARE DISEASES THIS IS HOW YOU HAVE TO DO IT. IF YOU WANT TO STUDY MULTIPLE FACTORS SIMULTANEOUSLY, AGAIN, THESE TYPES OF STUDIES ARE GOOD FOR THAT. YOU'RE COLLECTING A TON OF DATA TO TRY TO FIGURE OUT WHAT IS DIFFERENT BETWEEN GROUPS.
57:26
IF YOU KNEW THE ANSWER, IT WOULD BE A LOT EASIER, BUT YOU DON'T. YOU'RE TRYING TO COLLECT A LOT OF INFORMATION AND SEE. IT'S ACTUALLY NOT THAT TIME CONSUMING. IT'S LESS EXPENSIVE THAN SOME STUDIES. I'M NOT SAYING THAT IT MAY NOT TAKE YOU TEN YEARS. BUT IT MIGHT TAKE YOU LESS THAN A YEAR. THE WHOLE CLAUSE, IF THE ASSUMPTIONS ARE MET, WHICH IS HARD TO KNOW UNTIL AFTER YOU'RE DONE AND A LOT MORE WORK HAS BEEN DONE, THE ASSOCIATIONS AND RISK ESTIMATES THAT I SEE IN A REALLY WELL DONE CASE CONTROL
58:01
STUDY CAN BE VERY CONSISTENT WITH OTHER TYPES OF STUDIES. SO WE'VE DONE CASE CONTROL AND THEN LATER WE'RE ABLE TO DO LATER ADDITIONAL RANDOMIZED EXPERIMENTS AND FOLLOW-ON EXPERIMENTS. IF YOU HAVE A PRETTY REPRESENTATIVE SAMPLE OF BOTH YOUR CASES AND CONTROLS, YOU GET A VERY GOOD ESTIMATE BUT YOU'RE NEVER GOING TO KNOW THAT UNTIL YOU'VE DONE THE OTHER STUDIES. NEGATIVE ATTRIBUTES YOU DON'T ESTIMATE INCIDENCE. WHY? SET THE NUMBER OF CASES IN MY TRIAL. I DON'T KNOW THE INCIDENCES.
58:34
I DON'T ESTIMATE PREVALENCE FOR THE EXACT SAME REASON. THE RELATIVE RISK IS INDIRECTLY MEASURED, SO I MENTIONED IT THERE, BUT IF YOU HAVE A RARE DISEASE, SO WHEN YOU DO CASE CONTROL STUDIES YOU CAN ONLY TECHNICALLY DIRECTLY MEASURE THE 0S RATIO. BUT YOU CAN GET AN SYSTEM OF THE RELATIVE RISK. WE TALK ABOUT THAT IN THE TEXTBOOK. BIAS IS AN ISSUE. BIAS IS AN ISSUE IN EVERY STUDY. YOU DO METHODS TO TRY TO MINIMIZE YOUR BIAS. IT'S HARD TO STUDY A RARE EXPOSURE. SO REMEMBER, I CAN STUDY RARE
59:11
DISEASES USING THIS BUT A RARE EXPOSURE, I NEED TO USE COHORT STUDIES FOR. AND I PROBABLY WANT TO SAMPLE ON THE EXPOSURE. SO WHATEVER IS RARE, I TRY TO USUALLY SAMPLE ON IF POSSIBLE. ANOTHER NEGATIVE ATTRIBUTE OF THESE CASE CONTROL STUDIES IS THE TEMP PORAL RELATIONSHIP IS VERY HARD TO DOCUMENT A LOT OF TIMES PEOPLE ARE REMEMBERING RETROSPECTIVELY WHAT HAPPENED. IT'S HARD TO FIGURE OUT THE CAUSE AND THE EFFECT. BUT EXAMPLES WHERE THIS WAS DONE PRETTY WELL, THIS IS A CASE
59:45
CONTROL DESIGN BACK IN THE 1974 BUT I THINK IT'S REALLY COOL. I KNOW -- ACTUALLY ONE REASON I USE THE OLDER CITIES, PEOPLE DON'T -- STUDIES, PEOPLE DON'T SUE ME AS MUCH. THE OTHER REASON IS SOME OF THOSE ARE LIKE GOOD CLASSIC EXAMPLES. THIS CASE CONTROL DECIDE IDENTIFY THE RELATIONSHIP BETWEEN THE EXPOSURE TO A DRUG DURING A MOTHER'S PREGNANCY, OCCURRENCE OF A RARE DO YOU REMEMBER IN THE FEMALE OFF SPRING MANY YEARS LATER. I MEAN LIKE THEY WERE 10-18 YEARS OLD. HOW THE HECK DO YOU FIGURE THAT
1:00:18
OUT? THIS IS THE TYPE OF STUFF THAT RALLY WELL DONE CASE CONTROL CITIES, THIS IS THE TYPE OF ANSWER THAT YOU SEE. WHEN YOU START SEEING THIS VERY RARE TYPE OF VAGINAL CANCERS SHOW UP IN MUCH YOUNGER FOLKS THAN YOU EXPECT. YOU SEE A CASE SERIES THAT GETS PUBLISHED. AND THEN YOU START TO FIND ALL THESE DIFFERENT CASES AROUND THE WORLD. YOU TRY TO FIND MATCHED CONTROLS. YOU GO THROUGH ALL THE LITTLE GIRLS EXPOSURES. THEN WHEN YOU SEE NOTHING YOU START ASKING MOM. YOU SIGH WHAT'S GOING ON?
1:00:55
BUT MOM -- WHEN YOU HAVE A 10-YARD, MOM IS THERE ALREADY. IT'S HARD TO REMEMBER, DO YOU REMEMBER THE DRUGS YOU TOOK LAST WEEK? LET ALONE WHAT YOU MAY HAVE TAKEN TEN TO 20 YEARS AGO? BUT THIS ACTUALLY HELD UP OVER TIME. YOU HAVE ANOTHER ONE, THIS IS ONE THAT JERRY MENCALF SPEAKING TO YOU INGUEN WORKED ON. SO ONCE -- TO YOU WORKED ON. ONCE I NEEDED SOMEONE TO COVER ME. JERRY IS LIKE I'VE DONE A EPI STUDY. HE DID THE RISK FACTORS FOR THOUGH POST OPERATIVE [INDISCERNIBLE], BUT BASICALLY,
1:01:37
IT WAS THE LINK BETWEEN THISNER OK LURE LENS MATERIAL AND THE SERIOUS INFECTION FOLLOWING CATARACT SURGERY. A LOT OF DIMES YOU'LL -- TIMES YOU'LL SEE A MEDICAL ASKED OR FELLOW TRYING TO DO A STUDY. EVERY ONE SAID -- HE TELLS THE STORY BETTER, BUT HE SAID THE MATERIAL COULDN'T MATTER. WHAT ARE YOU TALKING ABOUT, AND -- NO, THE TYPE OF LENS THAT YOU USE, THE MATERIAL IN IT, IS VERY CLOSELY DO ID. SO HE FIGURED IT OUT BECAUSE OF THE MANUFACTURER THAT WAS LISTED IN THE MEDICAL RECORDS FOR THESE
1:02:15
DIFFERENT SURGERIES. AND IT ENDED UP WITH A LOT OF MATERIAL AND ACTUALLY A BIG CHANGE IN MANUFACTURING AROUND THE WORLD. FOR CATARACT SURGERY LENSES. NOW WE'RE GOING TO MOVE ON TO COHORT STUDIES. PERSPECTIVE OR LONGITUDINAL COHORT STUDIES ARE YOUR MOVIES. I TEND TO TAKE OBSERVATIONS CONCERNING ASSOCIATIONS BETWEEN SOME EXPOSURE OR RISK FACTOR, AND SUBSEQUENT DEVELOPMENT OF DISEASE. I MIGHT EXAMINE MULTIPLE OUTCOMES FOR A SINGLE EXPOSURE. SO WHAT'S A SINGLE EXPOSURE?
1:03:04
SOME FOLKS ONLINE WERE TALKING ABOUT LIBYA'S EBOLA SURVIVERS NATURAL HISTORY STUDY. I HAVE TO ADMIT I KNOW NEXT TO NOTHING ABOUT THIS STUDY. ALL RIGHT? SO I COULD GET THE DETAILS WRONG IN WHICH CASE CLASSMATES TYPE IT OUT FOR ME ONLINE. BECAUSE PEOPLE STARTED TALKING ABOUT IT ONLINE TODAY, SO AS I WAS IN BETWEEN IMMEDIATINGS I WAS -- MEETINGS, I WAS TRYING TO LOOK UP MORE AND MORE INFORMATION. YOU COULD SAY IN THIS CASE, AND I'M NOT SURE FITS AA LONGITUDINAL DECIDE OR NOT. I THINK I SAID.
1:03:35
BUT I WASN'T 100% SURE FROM WHAT I READ ONLINE. FOR EXAMPLE, I'M GOING TO MAKE IT A LONGITUDINAL STUDY FOR MY EXAMPLE WHETHER IT IS OR ISN'T. PEOPLE HAVE BEEN EXPOTIONED TO THE EBOLA VIRUSES. THEY HAVE SURVIVED. I DON'T KNOW WHAT'S GOING TO HAPPEN TO THEM OF WARDS. WE'RE STARTING TO SEE PROBLEMS WITH THEIR VISION. WE DON'T KNOW HOW LONG THE VIRUS IS GOING TO SURVIVE. ARE THEY PASSING IT ON TO THEIR CAPITAL PARTNERS AND FOR -- SEXUAL PARTNERS AND FOR HOW LONG? THERE ARE LOT OF DIFFERENT
1:04:10
OUTCOMES WOE MAY BE INTERESTED IN. WHEN THE WORLD TRADE CENTERS COLLAPSED, CDC STARTED A STUDY OF THE PEOPLE WHO WERE DOING CLEANUP. THEY SAID ALL RIGHT, EASY THING WE WANT TO MEASURE. RESPIRATORY DISTRESS AND POST TRAUMATIC STRESS. BUT THEY MEASURED A LOT OF OTHER THINGS ABOUT THESE PEOPLE. THEY KEPT THEIR BLOOD SAMPLES, THEY ASKED LOTS OF QUESTIONS, AND HAD BEEN DOING THIS OVER TIME. WHY? BECAUSE WE DON'T KNOW WHAT THAT EXPOSURE IS GOING TO CAUSE AT THE END. SO WHAT THESE LONGITUDINAL
1:04:44
COHORT STUDIES ARE GOOD FOR, TO TRY TO MEASURE A LOT OF DIFFERENT POTENTIAL OUTCOMES FROM A SINGLE EXPOSURE HAVE THEY'RE ALSO GOOD TO FIGURE OUT THE NATURAL HISTORY OF A POPULATION. SO WE DID A STUDY LIKE THIS IN PEOPLE WHO WERE 65 AND OLDER IN THE UNITED STATES. IT'S CALLED THE CARDIOVASCULAR HEALTH STUDY. THEY ENROLLED PEOPLE AT CERTAIN SITES AROUND THE UNITED STATES THAT WERE ON THE MEDICARE ROLES. AND THIS STUDY RAN FOR 10 YEARS. THEY SAID WE WANT TO KNOW HOW AMERICANS ARE AGING IN CURRENT
1:05:20
SOCIETY. WHAT IS HAPPENING? AND THAT'S WHAT'S USEFUL ABOUT THESE STUDIES. US DIRECTLY CALCULATE THE INCIDENCE OF DISEASE FOR EACH EXPOSURE GROUP IN A STUDY LIKE THIS. YOU COULD DO A LOT OF OTHER REALLY COOL WORK, TOO. SO IT'S NOT NECESSARILY ALWAYS GOING TO SHOW YOU HOW TO TREAT SOMEBODY BUT IT DOES GIVE YOU AN IDEA KIND OF WHAT'S THE EVIDENCE. AND WHAT'S THE UNDERLYING INFORMATION. SO HERE IS YOUR PICTURE. 2 WEES -- WAYS THAT THIS HAPPENS. CONCURRENT. SO TODAY I DEFINE THE POPULATION
1:06:01
AND THEIR EXPOSURE STATUS. I'M GOING TO FOLLOW YOU UNTIL 2025. TEN YEARS. I'M GOING TO SEE, DO YOU HAVE THE DISEASE OR NOT? IF YOU WERE EXPOSED TO THE EBOLA VIRUS, IF YOU WEREN'T. IF YOU'RE EXPOSED TO WORKING ON THE WORLD TRADE CENTER SITE, IF YOU WERE NOT EXPOSED TO THAT. OR I DO WHAT'S CALLED A NON CONCURRENT HISTORICAL STUDY. TODAY I SEE A GROUP OF PEOPLE THAT OF A DISEASE OR DON'T HAVE A DISEASE. AND I TRY TO SAY WHAT WAS YOUR EXPOSURE STATUS BACK IN 1980? SO I'M STILL DEFINING A
1:06:44
POPULATION BY THEIR EXPOSURE STATUS BUT IT'S IN THE PAST. HOW DO YOU FIGURE THIS OUT? USUALLY FROM WORK RECORDS. SO CLASSIC EXAMPLE IS A NICKEL REFINERY STUDY. I THINK WAS PUBLISHED BY LEON GORDIS AND HIS EPIDEMIOLOGY TEXTBOOK WHERE THEY BASICALLY LOOKED BACK AT WHO WORKED AT THE NICKEL REFINERY, AND WOULD HAVE BEEN EXPOSED OR NOT EXPOSED TO THE CERTAIN TOXIC CHEMICAL. 50 YEARS LATER, DID THEY HAVE CANCER. I SEE THIS ALSO, THE DEPARTMENT OF DEFENSE HAS DONE SEVERAL STUDIES LIKE THIS.
1:07:20
THEY LOOK BACK AT RECORDS FROM PEOPLE WHO WERE THERE IN WORLD WAR II. AND WHAT TYPES OF EXPOSURES THEY HAD. SO LET'S GO THROUGH THE CONCURRENT PROSPECTIVE STUDIES FIRST. YOU IDENTIFY A POPULATION THAT YOU WANT TO SURVEY. YOU DEFINE THAT. IF YOU HAVE RISK FACTORS IN MIND, WHICH NOT ALL OF THESE STUDIES DO. CARDIOVASCULAR HEALTH DECIDE, YOU WERE 65 OR OLDER ON MEDICARE IN THE U.S. DONE. THEY HAD NO OTHER PREDEFINED RISK FACTORS. THEY JUST MEASURED A TON OF STUFF. IF YOU HAVE AND IDEA OF A
1:07:57
PARTICULAR RISK FACTOR EXPOSURE, AGAIN, YOU HAVE TO IDENTIFY IT, IT HAS TO BE CLEAR. YOU NEED TO IDENTIFY PEOPLE WITH IT AND WITHOUT IT. YOU'RE GOING TO FOLLOW THEM FORWARD IN TIME AND COMPARE THE INCIDENCE RATE OF ALL THESE DIFFERENT DISEASES BETWEEN THE GROUPS. IF IT'S RETROSPECTIVE, YOU NEED TO DEFINE THE POPULATION WITH THE PRESENCE OR ABSENCE OF THE EXEXPOSURE, AND YOU'VE GOT TO DO THAT IN AN ACCURATE OBJECTIVE FASHION. THAT'S WHERE THOSE EMPLOYMENT RECORDS COME IN.
1:08:29
THE DATA NEEDS TO BE COLLECTED PROSPECTIVELY ON THE DISEASE, HOWEVER. THE PROBLEM WITH RETROSPECTIVE STUDIES, PEOPLE ARE RECALLING INFORMATION. THEY MAY HAVE VERY DIFFERENT RECALL. WE TRY TO DO THIS FOR SOME HERBICIDE DISEASED. WELL, DO YOU REALLY KNOW IF YOU WERE EXPOS CONGRATULATED DO YOU KNOW WHAT YOU WERE EXPOSED TO, HOW MUCH YOU HAD? IT WAS VERY HARD. TO TRY TO GET ACCURATE INFORMATION ABOUT EXPOSURE. BUT IF YOU'RE DOING THESE RETROSPECTIVE OR NON CONCURRENT STUDIES, THE IDEA THAT WHAT I'M
1:09:06
SURVEYING IN THE -- DO YOU HAVE THE DISEASE, YES OR NO. BUT I HAVE TO DEFINE THE INCIDENT RATES FOR THOSE EXPOSED OR NON EXPOSED. IF YOU THINK YOU KNOW WHAT YOU'RE EXPOSURE IS, YOU NEED TO MAKE SURE THAT YOU'RE EXPOSED AND NON EXPOSED ARE REPRESENTATIVE AND WELL DEFINED. THAT YOUR NON EXPOS RESIDENT TRULY NON EXPOSED. THIS CAN BE A PROBLEM. THE ABSENCE OF EXPOSURE, COULD BE HARDER TO DEFINE TO BE HONEST WITH YOU AND THE OTHER TRICK IS YOU HAVE TO ASSUME THAT PEOPLE MAINTAINED THEIR NON EXPOSURE
1:09:45
STATUS THROUGHOUT THE STUDY. SO IF YOU'RE GOING TO FOLLOW THEM FOR 20 YEARS, AND THEY SUDDENLY GET EXPOSED, YEAR 15, HOW IS THAT GOING TO IMPACT YOUR RESULTS? SOMETHING WITH TRAUMATIC BRAIN INJURY ISSUES, SO THERE IS A DOD STUDY WE'LL DISCUSS AT THE END OF THIS WHERE THEY LOOKED AT PEOPLE THAT HAD BEEN DIAGNOSED WITH DRAMATIC BRAIN INJURY DURING WORLD WAR T THEY'RE LOOKING TERRIFIC OF ALZHEIMER'S LATER IN LIFE. -- LOOKING TERRIFIC OF ALZHEIMER'S. THERE IS A LOT OF THINGS THROUGH
1:10:21
CAN CALL TRAUMATIC BRAIN INJURY. A BAD CAR WRECK CAN CAUSE THESE SAME TYPE OF INJURIES THAT WHAT THEY WERE DEFINING HAPPENED IN WORLD WAR II. SO WHAT DO YOU DO? DO I ASK PEOPLE IF THEY HAD THESE FOLLOWING INCIDENCES HAPPEN? OR INSTEAD, DO YOU GO THROUGH AND SAY, WELL, I'M JUST GOING TO ASSUME THEY DIDN'T. WHICH MEANS THAT YOU PROBABLY HAVE A LOT MORE VARIABILITY IN THOSE POPULATIONS. ALL OF THE PROBLEMS THAT I TALK ABOUT CAN LEAD TO INCORRECT ESTIMATIONS INCORRECT INTERPRETATIONS.
1:11:05
DOESN'T ALWAYS MEAN THAT THE STUDY IS WRONG. IT DOES MEAN AS YOU'RE READING THE STUDY AND YOU'RE THE PERSON THAT IS ANALYZING IT AND WRITING THE STUDY REPORTS, YOU NEED TO SIDDHIS ARE LIMITATIONS THAT YOU HAVE. YOU ALSO HOPEFULLY CAN SAY I'M GOING TO TRY TO COLLECT OTHER INFORMATION AND DO WHAT'S CALLED THE SENSITIVITY ANALYSIS TO TRY TO FIGURE OUT KIND OF HOW WRONG MY DATA COULD BE. HOW EXTREME DO I HAVE TO BE OFF BEFORE MY STUDY RESULTS CHANGE? AND I TELL THE DIFFERENT STORY. BACK TO THIS.
1:11:41
YOUR OUTCOME. WHAT IS THE DISEASE? DOWS DISEASE DEFINE EVERYTHING ELSE IS HARD OR RATHER NOT HARD. IT'S SUBJECTIVE. SO WHEN DOES YOUR CANCER GET DIAGNOSED? THAT DEPENDS A LOT ON WHERE YOU LIVE, HOW MUCH MONEY YOU HAVE. IF I DO OR DO NOT HAVE INSURANCE AND HOW GOOD IT IS. IT DEPENDS ON CULTURE. SOME PLACES I HAVE BEEN FOR BREAST CANCER, THESE WOMEN, IT IS BURSTING OUT OF THEIR BREASTS. BUT THEY WANT TO IGNORE IT. YOU HAVE TO DEFINE THE DISEASE OUTCOME. AND YOU'RE GOING TO PROBABLY
1:12:21
HAVE TO BUILD INTO YOUR BUDGET THAT YOU'RE SCREENING FOR THE DISEASE. IN ORDER 250 MAKE SURE THAT YOU HAVE CONSISTENT OUTCOME INFORMATION. YOUR PROSPECTIVE LONGITUDINAL COHORT STUDIES, THE SAME CRITERIA ARE APPLIED. TO THE EXPOSED AND NON EXPOSED GROUPS. I KEEP SAYS THIS OVER AND OVER. DEFINITIONS OF DISEASE HAVE TO BE RELIABLE AND REPRODUCIBILITY. SOMETIMES IF YOU'RE DOING THAT PROSPECTIVE STUDY, THE DEFINITION OF DISEASE MAY CHANGE DURING THE COURSE OF YOUR STUDY. WHAT THE WORLD THINKS THE
1:12:56
DEFINITION IS OR HOW YOU GO ABOUT CHECKING TO SEE AND DIAGNOSING A DISEASE MAY CHANGE. IN THAT CASE YOU HAVE TO UPDATE YOUR PROTOCOL AND YOU'LL HAVE TO SAY OKAY, HERE ARE MY ORIGINAL DISEASE DEFINITIONS, WHAT'S MY ANALYSIS WITH THAT. >> HERE IS THE NEW GUIDELINE DEFINITIONS THAT CAME OUT TIN MAINTAIN. WHAT'S MY ANALYSIS IF I USE THOSE? YOU HAVE TO HANDLE THEM. YOU HAVE 250 MINIMIZE YOUR LOSS OF FOLLOW UP. IF YOU HAVE A NON RESPONSE RATE GREATER THAN 20%, IT'S GOING TO RAISE QUESTIONS ABOUT THE
1:13:27
ACCURACY OF YOUR INCIDENCE RATES. SO IF YOU HAVE -- THEY'RE LOOKING FOR A CERTAIN NUMBER OF FOLKS THAT HAD EBOLA AND CERTAIN NUMBER OF CLOSE ENCOUNTERS, IN THAT NATURAL HISTORY STUDY, IF THEY CAN'T GET THOSE NUMBERS, WHY? WHY ARE PEOPLE NOT SIGNING UP FOR THEIR STUDY? IF THEY ONLY DO IT IN ONE SPECIFIC HOSPITAL REGION, THAT'S USEFUL. BUT IS IT AS USEFUL AS GENERALIZABLE IF THEY MOVE ACROSS TO OTHER GROUPS? SO CALCULATE THE INCIDENCE FOR THE STUDY PERIOD. AGAIN, YOU START TO SEE ALL
1:14:06
THESE TESTS, BUT REALLY, YOU SHOULD BE USING REGRESSION. A POSITIVE ATTRIBUTE OF THOSE TYPES OF STUDIES. THEY'RE A LOT MORE REPRESENTATIVE OF CASES THAN A CASE CONTROL STUDY. WHY I'M PICKING UP MY INCIDENT CASES NOW. YOU GET A LOT MORE NATURAL HISTORY INFORMATION WHEN YOU DO THESE STUDIES. YOU WATCH THE COURSE OF DISEASE ALL ALL THE AFTERMATH AFTER DATION. IT'S A BEAUTIFUL TYPE OF DECIDE TO DO. THE COHORT STUDIES HAVE BEEN THE MORE PLEASURABLE EXPERIENCES BECAUSE YOU FEEL LIKE YOU'RE
1:14:44
ABLE TO ACTUALLY GIVE INFORMATION TO NEW PATIENTS THAT THEY DIDN'T HAVE BEFORE. INCIDENCE RATES ARE INVIABLE IF YOU'VE DONE YOUR JOB WELL. YOU HAVE REPRESENTATIVE GROUPS. AND WE CAN DIRECTLY ESTIMATE THE RELATIVE RISK. SO AT THE END OF TODAY'S LECTURE, YOU'LL HAVE FROM SLIDES SOME 5 TO 131, THESE ARE TRACT -- 105 TO 131. THESE ARE SLIDES FOR YOUR TO REED. SO YOU CAN DIG INTO THEM. SO WE'RE NOT GOING TO DISCUSS THEM TONIGHT. HE TALKS ABOUT HOW PEOPLE BASICALLY MORE NATURALLY
1:15:23
INTERPRET RELATIVE RISK TO ODDS RATIOS. I THINK PAUL WILL TALK ABOUT THAT IN HIS HYPOTHESIS TESTING. IF YOU WANT ME TO COVER IT MORE IN THE SUMMARY LECTURE, I CAN DO THAT TOO. BUT IT'S VERY TRICKY ELEMENT TO THINK ABOUT. SO THE POSITIVE ATTRIBUTES, THIS IS A LOT LESS BIAS COMPARED TO OTHER STUDY TYPES. THEY CAN BE VERY BIASED ALSO, DON'T GET ME WRONG. YOU SEE A RELATIONSHIP TO TEXT POSHER AND YOU CAN ACTUALLY PICK UP A TEMP PORAL RELATIONSHIP, MUCH EASIER PROSPECTIVELY THAN
1:15:58
HISTORICALLY BUT YOU CAN PICK IT UP. RATHER YOU'RE MORE LIKELY TO. RARE EXPOSURE WITH FREQUENT CASES AMONG THE EXPOSED SHOW UP BEAUTIFULLY HERE. THE PROBLEMATIC ATTRIBUTE, YOU MAY NEED REALLY, REALLY, REALLY LONG FOLLOW-UP. IT MIGHT TAKE 50 YEARS FOR DEMENTIA TO SHOW UP. THE FREE LIVING POPULATION FOLLOW-UP IS EXPENSIVE. IT'S HARD TO TRACK DOWN THESE PEOPLE. AND YOU KEEP TRACKING THEM DOWN EVERY TIME YOU WANT TO MEASURE THEM. A LARGE POPULATION IS USUALLY REQUIRED TO DO THESE ANALYSES
1:16:34
AND HAVE REASONABLE POWER. AND REASONABLE BELIEF THAT YOU'RE ACTUALLY HAVE GENERALIZABLE RESULTS. YOU NEED BASELINE DATA. IF YOU DON'T GET INFORMATION ABOUT PEOPLE CLOSE TO THEIR EXPOSURE TIME, YOU CAN RUN INTO A LOT OF PROBLEMS. A RARE DISEASE REALLY CANNOT BE STUDIED IN THIS METHOD. YOU DON'T KNOW IF YOU HAVE A BIG ENOUGH SAMPLE SIZE TO PICK UP ENOUGH PEOPLE WITH THE RARE DISEASE. BUT YOU CAN GET RARE EXPOSURES. SO GUFF A RARE EXPOSHer, USE THE COHORT STUDY. IF YOU JUST WANT TO FOLLOW A
1:17:10
BUNCH OF PEOPLE AND SEE WHAT'S HAPPENING DO A COHORT STUDY. IF YOU HAVE A RARE DISEASE, DO A CASE CONTROL STUDY. YOU CAN SEE A LOT OF BIAS, SO IT MAY BE LESS BIAS IN SOME THINGS, YOU SEE A LOT OF LOSS OF FOLLOW-UP ASSESSMENTS PROBLEMS WITH HOW THEY DO THE ASSESSMENTS BASED ON THE EXPOSURE, YOU'LL SEE ISSUES. AGAIN, HERE, I KNOW YOU HAD THIS DISEASE OR I KNOW YOU HAVE THIS EXPOSURE. I'M MORE LIKELY TO POKE AT YOU. ASK YOU AGAIN, ARE YOU SURE. SO I MENTIONED SOME OF THESE STUDIES.
1:17:46
THERE WAS A PROSPECTIVE COHORT STUDY THAT SHOWED AN EARLY INCREASE IN THE RISK OF LUNG CANCER AND THERE WAS A STUDY, ORIGINALLY OF 54. THEN 50 YEAR FOLLOW-UP ON THESE FOLKS. A GROUP OF MALE DOCTORS AND THEIR SMOKING HABITS. AND THEY FOLLOWED UP. BMJ LITERALLY 50 YEARS LATER. THEN THE MEDICAL RECORDS THAT I TALKED ABOUT, THIS IS THE STUDY ON THAT LOOKING AT ALZHEIMER'S DISEASE AND OTHER DEMENTIAS. AND NATURAL HEALTH INSURANCE RESEARCH DATABASE. THIS IS A STUDY THAT CAME FROM
1:18:31
ONE OF THE COURSE PARTICIPANTS A YEAR OR TWO AGO. HE ASKED ONLINE, THEN VIRUS INFECTION ASSOCIATED WITH INCREASED RISK OF CHILDHOOD TYPE WITHIN DIABETES IN TAIWAN. THIS IS THE QUESTION FOR YOU ALL TO ANSWER ONLINE. TELL ME WHAT THE STUDY DESIGN IS. WHEN YOU READ THROUGH THIS. AND I THINK ONE OF THE PDFs YOU HAVE IS ACTUALLY THE WRITEUP ON THIS STUDY. HERE IS THE ARTICLE IF YOU WANT TO LINK. THE LINK WAS LIVE AT LEAST ON FRIDAY WHEN I CHECKED IT. ARE THERE INTERIM OR EXPERSONAL -- INTERNAL OR
1:19:11
EXTERNAL VALIDITY ISSUES WITH THIS STUDY. THAT'S ANOTHER QUESTION TO ASK. THIS IS THE STUDY I MENTIONED ON POST TRAUMATIC STRESS SYMPTOMS AND ASTHMA, LOOKING AT EXPOSURE TO THE WORLD TRADE CENTER ATTACKS. SO NOT ALL COHORTS, THOUGH, ARE EXPOSURE BASED. I REALIZED GOING THROUGH THE LECTURE AND I WAS READING IT TO MYSELF AND I'M LIKE YOU KNOW, THEY'RE GOING TO THINK EVERYTHING IS ABOUT EXPOSURE BUT IT'S NOT. THIS IS A STUDY ALSO FROM ONE OF YOUR COURSE PARTICIPANTS ON VEGETARIAN DIETS AND THE
1:19:45
INCIDENCE OF CANCER. SO THIS IS HEALTH STUDY 2. THEY HAD CHOSE TO 70,000 PARTICIPANTS, LOOKING AT CANCER CASES. I MENTIONED EARLIER, A LITTLE BIT ABOUT NESTED STUDIES. SO DO REALIZE THAT YOU MIGHT NEST A CASE CONTROL OR CASE COHORT STUDY IN SIDE ONE OF THESE COHORT STUDIES. AN EXAMPLE OF A NESTED CASE CONTROL IS FROM YOUR PERSPECTIVE COHORT. I HAVE ALL THESE ZORED SAMPLES, RIGHT? I HAVE BASELINE, FOLLOW-UP DATA, I CAN NOW -- A LOT OF TIMES WAIT 6 MONTHS OR A YEAR. THEN THEY'LL TAKE ALL NEWLY
1:20:25
OCCURRING CASES OR A SAMPLE OF THOSE. AND COMPARE THOSE TO MATCH YOUR UNMATCHED CONTROLS. WHAT'S NICE, WELL, I CAN GO BACK AND DO THAT. I MIGHT BE IN YEAR 5 OR 10 OF MY COHORT STUDY. I'VE FOLLOWED THOSE DROLLS FOR A WHILE. I KNOW THEY DIDN'T BECOME A CASE. THEY'RE STILL NON CASES. SO I KNOW THEY'RE A CONTROL. I KNOW THEY DON'T HAVE DISEASE. OTHIS IS REALLY EFFICIENT. ESPECIALLY IF YOU HAVE EXPENSIVE OR DIFFICULT LAB ASSAYS YOU WANT TO RUN. YOU STORE EVERYTHING AND THEN
1:20:59
YOU ONLY RUN THE ASSAYS ON A SAMPLE OF FOLKS. AND IT MIGHT BE NICE BECAUSE YOU CAN SEE, DO I HAVE ENOUGH CASES IN THE COHORT IN ORDER TO ACTUALLY BOTHER TO RUN THE CASE CONTROL STUDY. YOU ALSO DON'T HAVE AS MANY SELECTION AND DATA COLLECTION ISSUES. YOU CAN STILL HAVE THEM, BUT THERE MAY BE FEWER. IN A CASE COHORT STUDY YOU'RE SELECTING FROM THE COHORT, ALL INCIDENT CASES AND COMPARING TO A RANDOM SUBSET OF NON CASES. AND YOU CAN READ ABOUT THIS MORE IN THE TEXTBOOK, OUR ADVANCED CHAPTER.
1:21:33
WE GO MORE INTO THIS. THESE ARE DIFFICULT TO ANALYZE. IN GENERAL YOU HAVE A LOT OF DIFFERENT TYPES OF OBSERVATIONAL STUDIES. THEY FALL INTO THESE CATEGORIES, CASE REPORTS, CASE SERIES, CROSS SECTIONAL SURVEYS, NATIONAL HEALTH INTERVIEW SURVEYS, ANOTHER COMMON SURVEY YOU DEDATA FROM. INSIDE THE UNITED STATES. CASE CONTROL STUDIES, GROUPS WITH OR WITHOUT THE DISEASE OUTCOME. AND DETERMINE WHO IS EXPOSED TO DIFFERENT RISK FACTORS, OR COHORT STUDIES WHERE YOU FOLLOW A GROUP FOR A WHILE OR YOU'RE
1:22:04
GOING TO CHOOSE PEOPLE WITH OR WITHOUT AN EXPOSURE STATUS. SO NOW LET'S TALK ABOUT BIAS. SOME FOLKS ARE GOING TO TALK TO YOU ABOUT 20 DIFFERENT TYPES OF BIAS. I DON'T CARE ABOUT THE LABELS. OTHER PEOPLE DO. SO WHEN YOU TAKE THEIR COMMON COUNCIL OF EXAMS -- THEIR EXAMS CARE ABOUT THE LABELS. AS A STATISTICIAN, I CARE ABOUT THERE IS VALUING. THERE IS A TRUE VALUE AND VALUE YOU COLLECTED IN THE STUDY. THEY'RE NOT THE SAME. THAT'S BIAS. I DON'T PARTICULARLY CARE IF IT'S SELECTION BIAS OR
1:22:48
INTERVIEWER BIAS OR ANYBODY ELSE'S BIAS. THERE IS A PROBLEM BECAUSE I'M NOT GETTING THE CORRECT INFORMATION. LIKE I SAID, A LOT OF OTHER PEOPLE GAVE ABOUT THE EXACT NAMES. I DON'T. BIAS IS BIAS. I'LL GO THROUGH A FEW COMMON NAMES. SELECTION BIAS, CALLED PREVALENCE INCIDENCE BIAS. IF YOU'RE EXPOSED OR IMPACTED EARLY I MIGHT MISS YOU. WHY? THE FATAL EPISODE F I GO OUT TO SURVEY WHO HAS EBOLA RIGHT NOW, IF YOU ALREADY DIED YOU DON'T EBOLA. YOU'RE DEAD. IF I ASK, SOMETIMES PEOPLE DON'T
1:23:32
KNOW WHAT TO CALL IT. TRANSIENT EPISODES, THE SILENT CASES IF. IF YOU HAD A SILENT STROKE NOBODY KNOWS UNTIL YOU GO IN FOR SOMETHING ELSE AND THEY FIGURE IT OUT. CASES WHERE THE EVIDENCE OF EXPOSURE DISAPPEARS WITH DISEASE ONSET. HAPPENS MORE FREQUENTLY THAN YOU MIGHT EXPECT. NON RESPONDENT BIAS. PEOPLE WERE UNWILLING OR UNABLE TO RESPOND TO YOUR KIND SURVEY. YOU DON'T KNOW ABOUT THEM. WHAT I ALSO DON'T KNOW IS IF YOU WOULD HAVE HAD DIFFERENT EXPOSURES AND OUTCOMES, AND
1:24:06
PEOPLE THAT WERE RESPONDERS AND NON RESPONDERS. SO SOMETIMES I GET THIS DIAGNOSTIC SUSPICION, EXPOSURE SUSPICION, RECALL BIAS, FAMILY INFORMATION BIAS. AND THIS CAN GO IN A VARIETY OF DIFFERENT WAYS. AGAIN, LET'S SAY IN YOUR FAMILY THAT THIS DISEASE IS PROMPT MONTH. MAYBE AS A GENETIC ASPECT. SO IF YOU COME FROM A FAMILY THAT HAS HUNTINGTON'S DISEASE IN IT, YOU'RE PROBABLY GOING TO BE LOOKING FOR AND THINKING ABOUT SOME OF THE EARLY WAYS THAT YOU WOULD DIAGNOSE HUNTINGTON'S
1:24:39
DISEASE, BEYOND THE GENETIC TEST. AT THE SAME TIME -- I HAD AND AUNT FOR YEARS, I THOUGHT SHE DIED FROM A STROKE IN HER 40s. THAT'S WHAT MY DAD TOLD ME. WHEN I HAD SEVERE HEADACHES IN MY 30s, THAT WAS IN MY FAMILY HISTORY. MY DOCTOR DID A FULL WORK UP ON ME. THEN I FOUND OUT THAT MY AUNT HAD KILLED HERSELF. SHE COMMITTED SUICIDE. BUT THAT WAS CONSIDERED SO TABOO IN MY FATHER'S FAMILY, THEY LIED TO EVERYBODY ABOUT IT. SO SOMETIMES WE TALK ABOUT OUR HISTORIES AND SOMETIMES WE DON'T
1:25:18
TELL THE TRUTH ABOUT OUR HISTORIES. THOSE CAN CAUSE A LOT OFF BIAS. OF BIAS. SO IN CONCLUSION, OBSERVATIONAL STUDIES ARE REALLY USEFUL. THEY COULD BE YOUR ONLY ALTERNATIVE. I LOVE RANDOMIZATION. I REALLY LOVE RANDIZATION. I HAVE COME UP WITH RANDOMIZED TRIALS AND PEOPLE TOLD ME THERE WAS NO WAY TO RANDOMIZE AND WE FIGURED OUT A WAY TO RANDOMIZE THAT STUDY AND IT WORKED. BUT SOMETIMES IT IS NOT AN OATH ALTERNATIVE. I CANNOT RANDOMIZE SMOKING IN HUMANS. I CANNOT RANDOMIZE PEOPLE WHO
1:25:56
LONG TERM ANTI-RETROVIRAL THERAPY FOR HIV. IT MAY BE UNTHEY WILL OR MAYBE THEY'RE NOT GOING TO STICK TO IT FOR THAT LONG. TOO MANY CHANGES. WHAT AM I GOING TO PRETTY END OF THE DAY? IF YOU CANNOT INTERPRET THE RANDOMIZED STUDY YOU HAVE TO REALLY THINK ABOUT WHAT YOUR OTHER OPTIONS ARE. SOMETIMES MY QUESTIONS ARE NOT RANDOMIZABLE. CARDIOVASCULAR HEALTH STUDY. WHAT HAPPENS IN FREE LIVING PEOPLE? YOU DON'T RANDOMIZE THAT STUDY. SOMETIMES PEOPLE FALL BACK AND SAY WELL IT'S CHEAPER AND FASTER
1:26:29
THAN A TRIAL TO DO AN OBSERVATIONAL STUDY. THAT'S TRUE, TOO. BUT CHEAPER AND FASTER MAY NOT BE BETTER. YOU HAVE TO BALANCE WHAT YOU NEED. EVERY ONE IS LAUGHING. SERIOUSLY, WE HAVE GONE OFF THE RAILS ON DISEASED. YOU HAVE TO DO A GOOD JOB. IF YOU CAN'T DO A GOOD JOB, DON'T DO THE STUDY. OBSERVATIONAL STUDIES DO NOT ALWAYS AGREE WITH THEMSELVES AND THEY DO NOT ALWAYS AGREE WITH OUR RANDOMIZED RESEARCH. HORMONE REPLACEMENT THERAPY IS A VERY LONG EXAMPLE OF THAT. YOU
1:27:02
CAN READ ABOUT IT EXTENSIVELY OR WE CAN TALK ABOUT IT. BUT THERE IS ALWAYS MANY TIMES PUBLICATION BIAS. YOU HAVE TO TAKE THAT INTO ACCOUNT. INCORRECT ANALYSIS OF OBSERVATION 8 STUDIES IS ALSO A BIG ISSUE, ESPECIALLY IF YOU'RE LOOKING BACK IN TIME. SOMETIMES YOU HAVE DIFFERENT POPULATIONS. THE POPULATION THAT WAS IN THE WOMEN'S HEALTH INITIATIVE WAS VERY DIFFERENT THAN THE POPULATION TYPICALLY TAKING HORMONE REPLACEMENT THERAPY. THAT HAS TO BE TAKEN INTO ACCOUNT. YOU ALSO HAVE TO THINK ABOUT
1:27:32
WHAT THE POPULATIONS FOR YOUR EPI STUDIES AND RANDOMIZED TRIAL AND WHAT'S YOUR GENERAL POPULATION? OBSERVATIONAL STUDIES WILL FIND A WEAKER DEGREE OF CONNECTION. THAT'S WHAT THEY'RE THERE FOR. YOU CAN CORRECT FOR WHAT YOU KNOW BY NOT FOR WHAT YOU DON'T KNOW. BUT AGAIN, IF IT'S UNETHICAL TO DO THE RANDOMIZED STUDY YOU'RE JUST GOING TO OF TO DO A REALLY GOOD JOB OF YOUR OBSERVATIONAL ONE. YOU'RE GOING TO HAVE ALL THE DESIGN ISSUES. PLACEBO EFFECT. INVESTIGATOR AND PARTANT BIASES.
1:28:04
TREATMENT ASSIGNMENT AND ASSESSMENTS. THAT'S WHAT FREE LIVING IS, RIGHT, BASICALLY AN OPEN LABEL EXPERIMENT. YOU GOT REGRESSION TO THE MEAN. YOU'RE GOING TO SEE NATURAL REDUCTION IN DISEASE ACTIVITY OVER TIME. SAME PROBLEM IN OBSERVATIONAL STUDIES. GOT CAUSATION VERSES ASSOCIATION. YOUR CAUSATION CAN ONLY COME WHEN YOU HAVE SOME RANDOMIZED EXPERIMENTAL STUDY OR COLUMN TRIAL INVOLVED. YOUR ASSOCIATION CAN COME FROM A LOT OF DIFFERENT TYPES OF STUDIES. RANDOMIZED, AND NON RANDOMIZED.
1:28:41
SO, WHAT DO YOU NEED TO DO? MEASURE EVERYTHING YOU CAN. BUILD INVESTIGATE A ROT OF MODELS. BE HONEST. DON'T JUST PUBLISHING THE ONE THAT LOOKS GOOD. TEST THE MODELS ON A LOT OF DIFFERENT DATA. TRY TO REPLICATE INSTITUTIONAL A LOT OF DIFFERENT DATA SETS. TRY PROPENSITY SCORES, TONS OF METHODS. HAVE A REASON FOR YOUR METHODS. BUT YOU'RE DRYING TO GENERATE HYPOTHESIS. FROM THIS. THERE ARE GOING TO BE A LOT OF PROBLEMS THAT YOUR BIGGEST MISTAKE I'VE SEEN. DESIGNING A RANDOMIZED TRIAL.
1:29:15
THAT'S THE BEST EVIDENCE WHEN YOU REALLY NEED AN OBSERVATIONAL STUDY. LOOK AT THE ALTERNATIVES, THINK ABOUT THE PREDECESSORS, YOUR FOLLOW UP STUDIES. THINK ABOUT POPULATION ASSESSMENT. DO YOU HAVE THE INFORMATION YOU NEED TO DO A RANDOMIZED DECIDE. >> GRANTED, YOU SHOULD BE WILLING TO DO CREATIVE RANDOMIZED STUDIES IF THAT'S WHAT YOU NEED. HERE IS SOME CASES WHERE I'VE BEEN ABLE TO RANDOMIZE BUT A LOT OF TIMES YOU CAN'T. WE CAN DISCUSS THOSE MORE ONLINE IF YOU WANT.
1:29:45
BUT YOU HAVE TO DEFEND YOUR NON RANDOMIZED STUDIES. RANDOMIZATION IS YOUR RESEARCH NORM. SOOGEED TO ADMITTED YOUR LIMITATIONS, AND DEFENDANT WHY I NEED TO AND HOW YOU'RE GOING TO MINIMIZE THE PROBLEMS THAT CAN COME UP. BUT DON'T FORGET THAT OBSERVATION IS YOUR FIRST STEP. FOR ANY RESEARCH. DON'T BE AFRAID TO LOOK. IF YOU WANT MORE INFORMATION ON PUBLISHING GUIDELINES THEY'RE PUBLISHED IN MANY DIFFERENT WAYS. EPI STUDIES HAVE THEIR OWN VERSION OF CONSORT. YOU HAVE A LOT OF SELF-GUIDED
1:30:16
COURSE INFORMATION HERE FROM THE CDC. AND THANK YOU AND IF YOU HAVE ANY ADDITIONAL QUESTIONS, PLEASE GO UHEAD AND SUBMIT THOSE ONLINE. WE HAVE SOME GREAT LECTURES TOMORROW TONIGHT AND I TO SEE YOU -- TONE NIGHT AND I HOPE TO SEE YOU IN THE COMING WEEKS, THANKS A LOT.
Here to help you learn
Sign in to keep a private conversation with your course tutor.
Ask me anything about the material…
Sign in to chat