Phil Hurvitz phurvitz@u.washington.edu EPI 511, Kukull Autumn, 2004 December 7, 2004 Assignment 7 1. The study used a case-control design. In four sentences or less answer the following: Was that the best design to answer their question? What are its strengths and weaknesses, compared to a cohort study, to look at this question? Case-control is an appropriate design for this study. In this instance it is far less expensive than a cohort study, can rely on existing data, and effective if schizophrenia and schizoaffective disorder are rare. In this specific study there is a problem that only a single measurement was taken per subject, which does not capture temporally varying levels of calcidiol (this is a threat to internal validity); a cohort study could be designed to take several samples per subject. Another general limitation of case-control studies is they do not allow direct calculation of incidence. 2. In five sentences or less answer the following: How were cases defined and how were they identified? Were these cases representative of the general population? Cases were identified through “through (a) personal interviews and/or (b) record link-age with psychiatric treatment facilities,” and by a follow-up interview following protocol in DSM-IV to find those defined to have schizophrenia and schizoaffective disorder. Cases could be representative of the general population under a few assumptions (i.e., the National Collaborative Perinatal Project participants are themselves representative of the general population; there was no bias in identification of cases). However, the very low sample size (< 0.2) is probably too low to give sufficient power for any conclusions regarding generalizability. In addition, the sample base includes only children of those mothers who were receiving prenatal care (not all D:\116108490.doc Hurvitz 1 of 4 mothers receive prenatal care, and there may be an association between receiving prenatal care and later development of psychological disorders). 3. In four sentences or less answer the following: How were controls selected? Was that method appropriate or would another control selection procedure have been better? “Controls were selected from an unaffected subset of subjects with no Axis I psychiatric diagnoses.” I would not consider this a random sample representative of the study base, as DSM-IV Axis I covers a large number of disorders (including cognitive, mood, anxiety, schizophrenia and other psychotic, substance-related, dissociative, sexual and gender identity, eating, sleep, and impulse-control disorders). How about just a random sample of those without schizophrenia and schizoaffective disorder? 4. In four sentences or less answer the following: A common disadvantage for case control studies is recall bias, is this an issue in this study? Why or why not? This is not an issue in this study, since the subjects or proxies are not being asked for their recollection of exposure; exposure was measured directly by serum test. However, there may be information bias due to potential problems identifying all cases based on records. 5. In two sentences or less answer the following: On page 76 we are given an OR of 0.98, (0.92-1.05). Interpret in words what the OR and the CI mean. The OR is close to 1, which means the odds of developing schizophrenia are no different among offspring of mothers who had either normal or low third trimester .calcidiol concentration. The confidence interval means that there is less than a 5% chance that the true OR is different than the calculated OR. D:\116108490.doc Hurvitz 2 of 4 6. In five sentences or less answer the following: What, if anything, does this study tell us about the incidence of schizophrenia? Explain your reasoning. To calculate incidence one must know the proportion of cases in the entire at-risk population (I assume this to be all births). In this study, cases were identified from a set of mothers who received prenatal care; this set is not representative of the entire at-risk population. 7. In four sentences or less answer the following: If we see no association between calcidiol levels during third trimester and schizophrenia in the offspring, what might account for this result if, hypothetically, a true association really DOES exist? What kind of error would occur to have this result? The authors admit that the long-term storage of samples may have lowered the amount of detectable calcidiol. This would “reduce the power to detect true but small group differences.” This would be an information bias. Another possibility is there is a confounding difference between the groups, such that the group with low calcidiol has some systematic behavior or exposure that is protective against psychological disorders, thereby nullifying the sole effect of low vitamin D. 8. In four sentences or less answer the following: What did the authors conclude from the study? Do you agree with them? Why or why not The authors conclude there is no evidence in this study linking prenatal vitamin D and schizophrenia in the offspring. I do not fully agree with their conclusion because of some of the problems I have listed above (non-random selection of controls, non-representative study base, possible degradation of samples over time). 9. You are an epidemiologist investigating the relation between indoor ETS (environmental tobacco smoke) and asthma. Upon review of your manuscript, one reviewer notes that your results are suspect because you did not control for daily pollen count, an exposure known to exacerbate asthma. Is this reviewer correct? Why or why not? This is not a complete description of the study and therefore cannot be answered straightforwardly. Is this an ecologic study? Cross-sectional?. Suppose this is a cohort study in D:\116108490.doc Hurvitz 3 of 4 which we can measure ETS exposure and asthma over time. If our population has the same pollen exposure (or for that matter, any other exposure that may affect asthma) within the ETS and non-exposed group, then the effect of the other exposure should be the same across both groups, so an increase in asthma can be ascribed to ETS. If there is differential exposure to pollen (perhaps rich people who can afford air conditioning tend to smoke less), then the criticism is warranted. 10. You are conducting a study of physical activity and risk of coronary heart disease (CHD). You enroll 2000 CHD-free men and women between the ages of 50 and 60 years and follow them up for 20 years, or until they develop CHD. Once per year, subjects visit your hospital clinic where among other things, their weight is assessed. You know that weight is related to both physical activity and CHD. Do you adjust for body weight in your analysis? In four sentences or less, explain why or why not. This is not a clear-cut decision. Overweight may actually be part of the causal chain (low activity → overweight → CHD) or vice versa (overweight → low activity → CHD), and as parts of causal chains, should not be considered confounders. low activity overweight CHD D:\116108490.doc Hurvitz 4 of 4