Epidemiology, Statistics Flashcards
You are designing a study to assess whether the risk of chorioamnionitis is higher after exposure to magnesium. You believe that exposed women will have a 33% higher rate of chorioamnionitis, but your colleague believes this number to be 50%. Holding all elements constant, what happens to the power as the effect size increases?
A. Increases
B. Decreases
C. Is unchanged
D. Is related only to sample size
A. Sample size, power, and effect size are related to each other. This simple formula demonstrates the relationship mathematically: n = [(Type I error) + (Power)]/effect size These numbers are generally fixed by the investigator, so the relationship between the sample size (n) and the effect size is clear. As the sample size increases, the detectable effect size decreases and vice versa. When the effect size is fixed, the power increases as the sample size increases. Finally, if the sample size is fixed, as the effect size increases, the power increases. Schulz KF, Grimes DA. The Lancet: Handbook of Essential Concepts in Clinical Research. Elsevier, Philadelphia, 2006.
- You wish to plan a study assessing whether the administration of supplemental vaginal progesterone impacts the time to delivery in twin gestations identified with a shortened cervix in the second trimester. Which is the best statistical test to assess your primary outcome (time to delivery) with adjustments for confounders?
A. Linear regression model
B. Logistic regression model
C. Ordinal regression model
D. Cox proportional hazards model
Answer:
D. The answer choices for this question are 4 types of regression models (linear, logistic, ordinal, and cox proportional hazards). In general, regressions are performed to assess how an exposure influences an outcome after adjusting for other confounders. The best answer to the question is the Cox proportional hazards model. This is a type of regression model that takes into account the time to an event – “time to delivery” in this case. While traditionally these models have been used in cancer research to assess time to death, the applications are excellent for obstetric research as well. In the case where one is assessing time to an event where this event is delivery, this model allows for variation in time of initiation of the treatment. For instance, not all of the twins in your sample had vaginal progesterone initiated at the same gestational age. It also allows one to account for “censoring,” the situation where either the event has not yet occurred, but the investigation period is over; or the patient has dropped out of the study. A linear regression is a regression model that adjusts for confounders when the outcome variable is continuous. For example, if you were assessing gestational age at delivery (assuming this is normally distributed), then linear regression would be most appropriate. Logistic regression is appropriate if your answer is categorical and dichotomous, generally yes/no. If your question were whether progesterone decreases the proportion of birth <37 weeks (ie, <37 wks yes/no), then logistic regression is your tool. Finally, ordinal regression is a special type of regression for categorical variables that are naturally ordered. If this question were rather a patient survey asking whether the patient felt progesterone lengthened her gestation and the answers ranged from “strongly agree” to “strongly disagree”, then ordinal regression is the most appropriate choice.
- A cohort study is planned to assess whether the latency from rupture to delivery is related to neonatal neurodevelopmental outcomes among pregnancies complicated by preterm premature ruptured membranes (PPROM). Which of the following comparisons most accurately identifies the exposure of interest?
A. Compare women with PPROM with women without PPROM for neonatal delays
B. Compare the neonatal neurodevelopmental outcomes among women with PPROM and a short latency versus those with a longer latency
C. Compare children with neurodevelopmental delay to children with without delay and assess the incidence of maternal PPROM in the two groups
D. Compare neonatal neurodevelopmental outcomes among women who deliver preterm after PPROM with women who deliver preterm after preterm labor with intact membranes
B. The principle tenet of the cohort study is the “EXPOSURE” of interest - participants are included and identified as “exposed” and “unexposed” to a hypothesized cause or factor relating to an outcome. Outcomes are then compared between the exposed and unexposed groups. In order to compile a correct cohort, one first has to ascertain the group at risk and all members of the at-risk group need to have the ability to develop the outcome. For example, in order to assess the incidence of malaria in pregnancy after using bed nets, your at-risk group should be pregnant women living in malaria-endemic areas rather than pregnant women in Washington DC. The women in Washington are much less likely to develop malaria. In this question, your cohort is defined as women with PPROM. Thus all included women should have this risk factor. Your exposure is the latency period, and your two comparison groups are women with a shortened latency compared with women with a longer latency. In these two groups, you will assess your outcome of neurodevelopment. Answer A is incorrect because women without PPROM will not have latency periods, thus you cannot assess your exposure. Answer C is incorrect because it represents a case-control design where participants are identified based on the outcome, in this case, neurodevelopmental delay. While answer D defines the groups a bit better than answer A, the intact membrane group will not have latency as defined by this question. Schulz KF, Grimes DA. The Lancet: Handbook of Essential Concepts in Clinical Research. Elsevier, Philadelphia, 2006.
Which of the following study designs cannot yield a relative risk?
A. Cohort
B. Case-control
C. Randomized clinical trial
D. Cross sectional study
B. Case-control studies are very efficient. This study design can be used with a rare “disease” or outcome, such as uterine rupture. By identifying a group with uterine rupture, then identifying a group with the same risk factors but no uterine rupture, one can then ascertain different exposures between the two groups. This design is efficient because it would take much longer to reach a rare outcome in a cohort study when participants are identified by exposure. Because case-control studies are pre-defined by the outcome, it is not possible to estimate the incidence of the outcome in these studies. Relative risk is measured by estimating the incidence of the outcome in exposed to the incidence of the outcome in non-exposed patients. Because the incidence cannot be calculated by an odds ratio, relative risk cannot be calculated by this study design. The odds ratio (the typical calculation for a case-control study) approaches the relative risk when the prevalence of the disease or outcome is low.
- A recent study showed an association between helicobacter pylori infection and preeclampsia. Which of the following supports a causal association?
A. Temporality, consistency, and biological plausibility
B. Temporality, deduction, and coherence with existing knowledge
C. Strength of association, counterfactual consideration, specificity
D. Temporality, analogy, duration
A. A statistically significant association does not imply causality. Several criteria have been proposed to establish causality; however, the most commonly used criteria were developed by Bradford Hill and published in 1965. In order for an exposure to be associated with the outcome, temporality must be established. In other words, the exposure must occur before the outcome. The strength of the association, as estimated by relative risk or odds ratio, should be taken into account. While a small association does not exclude causality, a stronger association is stronger evidence for this. The consistency of the association is also important. Have other researchers drawn similar conclusions? Is there a dose-response (biological gradient)? Causality could be inferred if a larger exposure leads to more of the outcome. The specificity of the outcome also implies causality. The exposure should only lead to the outcome with no other plausible explanation. Is there a biological plausibility explaining cause and effect? This is, of course, limited by the current knowledge. The other factors that suggest causality are coherence (is epidemiological evidence supported by laboratory findings?), experimental evidence (results of a randomized, controlled trial), or analogy (do similar factors have similar effects?). A deduction is inference and is not related to causality. The counterfactual is a concept in epidemiology that refers to a comparison group within a study. In order to examine whether there is a link between smoking and lung cancer, the best comparison would be made between the person who smoked and the same person if they rather did not smoke. In using the person’s counterfactual, one is controlling for all other confounders and only assessing the effect of the exposure on the outcome. Finally, duration only implies time but not causality. Schulz KF, Grimes DA. The Lancet: Handbook of Essential Concepts in Clinical Research. Elsevier, Philadelphia, 2006.
Hill, Austin Bradford (1965). “The Environment and Disease: Association or Causation?”. Proceedings of the Royal Society of Medicine 58 (5): 295–300.
You are interested in evaluating the levels of a new biomarker in women who have recurrent miscarriage copmared with parosu women without a history of miscarriage. Which of the following is the most appropriate study design?
A. Case-control
B. Cohort
C. Cross-sectional
D. Randomized trial
E. Survey
A. The correct answer is a case-control study. The type of data available and the patient population determine this conclusion. A cohort study would be more appropriate if the design were instead looking at a group of women who had known elevated levels of a certain biomarker, then following them through pregnancy and comparing how many had a viable birth vs. a pregnancy loss. A cohort study often requires long-term data collection (as in the NHANES and Framingham studies) and is designed to assess the effect of a defined exposure, without pre-defining the outcomes. The distinguishing feature of a cross-sectional study is that exposure and outcome data are collected at one specific point in time from a population (or representative subset); this data is often used to assess prevalence. Since data would not be collected in a single period of time for this study, a cross-sectional design is not appropriate. In this example, it is not possible to randomize women to have certain biomarker levels. In obstetric research, randomized trials are not common partly because the exposures of interest cannot be randomized, or because there is enough ethical concern about possible negative outcomes for the fetus that randomization would not be appropriate. Randomized trials are useful in determining superiority/non-inferiority of one intervention over another. A survey study would not be appropriate, since biomarker levels could not be adequately obtained through survey methodology. A case-control study would therefore answer this question most appropriately. This study design matches patients who have an outcome of interest (cases) with those who do not (controls) as well as other factors intended to eliminate sources of bias, then looks at the factors that contributed to those outcomes. Case-control studies are often ideal for the study of rare diseases (e.g. recurrent pregnancy loss).
. Which of the following sets of strategies includes only those recognized by the Society for Maternal-Fetal Medicine (SMFM) as strategies to facilitate safety and improve outcomes when using maternal-fetal medicine (MFM) services?
A. Frequent and clear communication between providers; delineating that obstetric ultrasound services should be provide by MFM Subspecialists
B. Encouraging the department or hospital to have an MFM subspecialist administratively over see labor and delivery; written documentation of each providers role and responsibilities
C. Encouraging policies that specify services to be rendered; encouraging the hospital or system to include an MFM subspecialist in their women and newborns’ administrative hierarchy
D. Encouraging development of specific policies and procedures regarding consultations, consultations with ongoing care of high risk conditions and transfer of care; clearly delineating the services to be rendered.
D. SMFM recognizes that certain strategies will facilitate safety and improve outcomes when using MFM services. These include:
-Encouraging specific policies and procedures regarding consultations, co-management of high-risk conditions, and transfer of care.
-Encouraging policies requiring written orders for MFM consultation.
-Encouraging policies that specify services requested [from MFM]. Clear delineation of services to be rendered [by obstetric providers and by MFM subspecialists].
-Written documentation of each provider’s role and responsibilities.
-Frequent and clear communication [between providers].
- What are the advantages of an ROC Plot?
A. Represents accuracy
B. It is dependent on prevalence
C. Cannot be used for decision “cut offs”
D. Cannot be used to generate confidence intervals for likelihood ratios
A.
The advantages of ROC curves include simple representation of the diagnostic test’s accuracy: The closer the curve is toward the upper left corner, the better the test is able to discriminate between disease and non-diseased state. ROC curves also allow determination of cutoff values that are clinically appropriate as tradeoffs between sensitivity and specificity. It also can be used to generate confidence intervals for sensitivity and specificity and likelihood ratios.
- What constitutes the axis of a receiver operator curve (ROC)?
A. X-axis: Sensitivity and Y-axis: specificity
B. X-axis: True positive rate and Y-axis: True negative rate
C. X-axis: True positive rate and Y-axis: false positive rate
D. X-axis: False positive rate and Y-axis: true positive rate
D.
In a ROC plot, the X-axis represents the false positive rate or 1-Specificity (specificity is basically the true negative rate). The Y-axis is sensitivity which true positive rate.
- Which of the following is true?
A. The larger the are under the curve (AUC) ther better
B. The smaller the AUC the better
C. AUC has no value in interpreting receiver operator curve analysis
D. An AUC of 0 indicates the prediction model is 100% right
A.
The advantages of an AUC include: 1) It is a measurement indicating how well predictions are ranked, rather than their absolute values. 2) It measures the quality of the model’s predictions irrespective of what classification threshold is chosen.
Robert Riffenberg, Statistics in Medicine third edition.
Bewick, Viv, Liz Cheek, and Jonathan Ball. “Statistics review 13: receiver operating characteristic curves.” Critical care 8.6 (2004): 508.
. Which of the following lists the correct variable to plot on the X-axis of an ROC curve?
A. Sensitivity
B. Specificity
C. sensitivity/specificity
B.
The correct answer is 1-specificity; this is the variable that should be plotted on the X axis. The corresponding variable on the Y axis is sensitivity.
ROC curves are a tool used in statistical analysis and in presentation of data. They are particularly useful in determining the utility of a diagnostic test or procedure, and for comparison of two different tests or procedures. A generic ROC curve is demonstrated in figure 1 below.
The reference line with a slope of 1 is indicative of a test whose results are no better than random chance in accurate diagnosis of a particular condition. The closer the apex of the curve is to the top left of the graph, the more accurate the test is. In order to compare two different tests to each other, the AUC or area under the curve must be calculated, and these numbers compared; an AUC of 0.5 is indicating that the test is equivalent to random chance in its ability to diagnose a disease, an AUC of 1.0 indicates that the test is perfect with no false positives or false negatives. A test with an AUC of 0 would be a test that always gives a false result, indicating that all disease positive samples have a negative result and all negative samples have a positive result.
In addition to comparing the performance of two different tests, another common use of ROC curves is determining the ideal cutoff point for a particular test.
which of the following is not used for power calculation?
A. Effect size
B. Alpha error
C. Type I errors (alpha)
D. Type II error (beta)
D.
In order to estimate power of an analysis of variance, you need the following:
-Number of treatment groups
-Sample size
-Risk of false positive you are willing to accept i.e alpha error
-Effect size you wish to detect (as known as delta)
A 26 year-old chemistry graduate student presents to her primary ob/gyn’s office for IUD removal as she is interested in getting pregnant. She has the ePSS (electronic Preventive Services Selector) application on her smart phone, a point of care tool promoted by the United States Preventive Services Task Force (USPSTF). She read that folic acid for the prevention of neural tube defects is a Grade A recommendation for women planning pregnancy. What is the definition of a Grade A USPSTF recommendation?
A. The USPSTF recommends the service. There is high certainty that the net benefit is substantial.
B. The USPSTF recommends the service. There is high certainty that the net benefit is moderate or there is moderate certainty that the net benefit is moderate to substantial.
C. The USPSTF recommends selectively offering or providing this service to individual patients based on professional judgment and patient preferences. There is at least moderate certainty that the net benefit is small.
D. The USPSTF recommends against the service. There is moderate or high certainty that the service has no net benefit or that the harms outweigh the benefits.
E. The USPSTF concludes that the current evidence is insufficient to assess the balance of benefits and harms of the service. Evidence is lacking, of poor quality, or conflicting, and the balance of benefits and harms cannot be determined.
A.
The USPSTF is an independent panel of experts in primary care and prevention who systematically reviews the evidence of effectiveness and develops recommendations for clinical preventive services. A summary of these recommendations is shown below.
Incidence cannot be calculated with which of the following study designs?
A. Randomized control trial
B. Cross sectional
C. Retrospective cohort
D. Prospective cohort
B.
Cross-sectional studies, also known as prevalence studies, examine the data on disease and exposure at one particular time point. Prevalent cases are all individuals living with the outcome of interest within a specified timeframe, regardless of when that person was diagnosed or developed the health outcome. Incident cases are all individuals who change in status from non-disease to disease – or from one state of a health outcome to another – over a specific period of time. In other words, “incidence” refers to the occurrence of new cases. Cross-sectional studies measure prevalent rather than incident cases.
Which of the following is true regarding increase in prevalence?
A. Positive predictive value increases
B. Negative predictive value increases
C. Increase in specificity
D. Increase in sensitivity
A.
As prevalence increases, the positive predictive value will go up and the negative predictive value will go down. Sensitivity and specificity have no relationship to prevalence.
