Robert W. Platt

Overadjustment bias and unnecessary adjustment in epidemiologic studies

Overadjustment is defined inconsistently. This term is meant to describe control (eg, by regression adjustment, stratification, or restriction) for a variable that either increases net bias or decreases precision without affecting bias. We define overadjustment bias as control for an intermediate variable (or a descending proxy for an intermediate variable) on a causal path from exposure to outcome. We define unnecessary adjustment as control for a variable that does not affect bias of the causal relation between exposure and outcome but may affect its precision. We use causal diagrams and an empirical example (the effect of maternal smoking on neonatal mortality) to illustrate and clarify the definition of overadjustment bias, and to distinguish overadjustment bias from unnecessary adjustment. Using simulations, we quantify the amount of bias associated with overadjustment. Moreover, we show that this bias is based on a different causal structure from confounding or selection biases. Overadjustment bias is not a finite sample bias, while inefficiencies due to control for unnecessary variables are a function of sample size.

Determinants of Preterm Birth Rates in Canada from 1981 through 1983 and from 1992 through 1994

BACKGROUND The rates of preterm birth have increased in many countries, including Canada, over the past 20 years. However, the factors underlying the increase are poorly understood. METHODS We used data from the Statistics Canada live-birth and stillbirth data bases to determine the effects of changes in the frequency of multiple births, registration of births occurring very early in gestation, patterns of obstetrical intervention, and use of ultrasonographic dating of gestational age on the rates of preterm birth in Canada from 1981 through 1983 and from 1992 through 1994. All births in 9 of the 12 provinces and territories of Canada were included. Logistic-regression analysis and Poisson regression analysis were used to estimate changes between the two three-year periods, after adjustment for the above-mentioned determinants of the likelihood of preterm births. RESULTS Preterm births increased from 6.3 percent of live births in 1981 through 1983 to 6.8 percent in 1992 through 1994, a relative increase of 9 percent (95 percent confidence interval, 7 to 10 percent). Among singleton births, preterm births increased by 5 percent (95 percent confidence interval, 3 to 6 percent). Multiple births increased from 1.9 percent to 2.1 percent of all live births; the rates of preterm birth among live births resulting from multiple gestations increased by 25 percent (95 percent confidence interval, 21 to 28 percent). Adjustment for the determinants of the likelihood of preterm birth reduced the increase in the rate of preterm birth to 3 percent among all live births and 1 percent among singleton births. CONCLUSIONS The recent increase in preterm births in Canada is largely attributable to changes in the frequency of multiple births, obstetrical intervention, and the use of ultrasound-based estimates of gestational age.

Genetic polymorphisms in methylenetetrahydrofolate reductase and methionine synthase, folate levels in red blood cells, and risk of neural tube defects

Folic acid administration to women in the periconceptional period reduces the occurrence of neural tube defects (NTDs) in their offspring. A polymorphism in the gene encoding methylenetetrahydrofolate reductase (MTHFR), 677C-->T, is the first genetic risk factor for NTDs in man identified at the molecular level. The gene encoding another folate-dependent enzyme, methionine synthase (MTR), has recently been cloned and a common variant, 2756A-->G, has been identified. We assessed genotypes and folate status in 56 patients with spina bifida, 62 mothers of patients, 97 children without NTDs (controls), and 90 mothers of controls, to determine the impact of these factors on NTD risk. Twenty percent of cases and 18% of case mothers were homozygous for the MTHFR polymorphism, compared to 11% of controls and 11% of control mothers, indicating that the mutant genotype conferred an increased risk for NTDs. The risk was further increased if both mother and child had this genotype. The MTR polymorphism was associated with a decreased O.R. (O.R.); none of the cases and only 10% of controls were homozygous for this variant. Red blood cell (RBC) folate was lower in cases and in case mothers, compared to their respective controls. Having a RBC folate in the lowest quartile of the control distribution was associated with an O.R. of 2.56 (95% CI 1.28-5.13) for being a case and of 3.05 (95% CI 1.54-6.03) for being a case mother. The combination of homozygous mutant MTHFR genotype and RBC folate in the lowest quartile conferred an O.R. for being a NTD case of 13.43 (CI 2.49-72.33) and an O.R. for having a child with NTD of 3.28 (CI 0.84-12.85). We propose that the genetic-nutrient interaction--MTHFR polymorphism and low folate status--is associated with a greater risk for NTDs than either variable alone.

Illustrating bias due to conditioning on a collider

That conditioning on a common effect of exposure and outcome may cause selection, or collider-stratification, bias is not intuitive. We provide two hypothetical examples to convey concepts underlying bias due to conditioning on a collider. In the first example, fever is a common effect of influenza and consumption of a tainted egg-salad sandwich. In the second example, case-status is a common effect of a genotype and an environmental factor. In both examples, conditioning on the common effect imparts an association between two otherwise independent variables; we call this selection bias.

Stress Pathways to Spontaneous Preterm Birth: The Role of Stressors, Psychological Distress, and Stress Hormones

The authors investigated a large number of stressors and measures of psychological distress in a multicenter, prospective cohort study of spontaneous preterm birth among 5,337 Montreal (Canada)-area women who delivered from October 1999 to April 2004. In addition, a nested case-control analysis (207 cases, 444 controls) was used to explore potential biologic pathways by analyzing maternal plasma corticotrophin-releasing hormone (CRH), placental histopathology, and (in a subset) maternal hair cortisol. Among the large number of stress and distress measures studied, only pregnancy-related anxiety was consistently and independently associated with spontaneous preterm birth (for values above the median, adjusted odds ratio = 1.8 (95% confidence interval: 1.3, 2.4)), with a dose-response relation across quartiles. The maternal plasma CRH concentration was significantly higher in cases than in controls in crude analyses but not after adjustment (for concentrations above the median, adjusted odds ratio = 1.1 (95% confidence interval: 0.8, 1.6)). In the subgroup (n = 117) of participants with a sufficient maternal hair sample, hair cortisol was positively associated with gestational age. Neither maternal plasma CRH, hair cortisol, nor placental histopathologic features of infection/inflammation, infarction, or maternal vasculopathy were significantly associated with pregnancy-related anxiety or any other stress or distress measure. The biologic pathways underlying stress-induced preterm birth remain poorly understood.

Effect of prolonged and exclusive breast feeding on risk of allergy and asthma: cluster randomised trial

Objective To assess whether exclusive and prolonged breast feeding reduces the risk of childhood asthma and allergy by age 6.5 years. Design Cluster randomised trial. Setting 31 Belarussian maternity hospitals and their affiliated polyclinics. Participants A total of 17 046 mother-infant pairs were enrolled, of whom 13 889 (81.5%) were followed up at age 6.5 years. Intervention Breastfeeding promotion intervention modelled on the WHO/UNICEF baby friendly hospital initiative. Main outcome measures International study of asthma and allergies in childhood (ISAAC) questionnaire and skin prick tests of five inhalant antigens. Results The experimental intervention led to a large increase in exclusive breast feeding at 3 months (44.3% v 6.4%; P<0.001) and a significantly higher prevalence of any breast feeding at all ages up to and including 12 months. The experimental group had no reduction in risks of allergic symptoms and diagnoses or positive skin prick tests. In fact, after exclusion of six sites (three experimental and three control) with suspiciously high rates of positive skin prick tests, risks were significantly increased in the experimental group for four of the five antigens. Conclusions These results do not support a protective effect of prolonged and exclusive breast feeding on asthma or allergy. Trial registration Current Controlled Trials ISRCTN37687716.

Should Meta-Analyses of Interventions Include Observational Studies in Addition to Randomized Controlled Trials? A Critical Examination of Underlying Principles

In their recent article, Shrier et al. (l) considered a 95% confidence interval as being one with a 95% probability that a population parameter is included in the interval and a 5% probability that it will lie outside the interval. We wish to clarify some aspects of confidence intervals’ interpretation. A simple example illustrates our points. Let us consider a group of 20 patients who receive a new treatment. Only 6 fail to respond to the treatment. Let p be the population proportion of possible patients who would not respond to the treatment. p is the parameter of interest; its true unknown value is the quantity to be estimated. Considering the Bernoulli process, the likelihood is p(1 p) for 0 < p < 1. We calculate an exact interval with 90% confidence (2) that happens to be nonsymmetric around 6/20: (0.175, 0.505). This may be the smallest (most precise) 90% confidence interval for the observation ‘‘6 out of 20.’’ The correct interpretation of the information that p is in this interval with 90% confidence is as follows: If we could repeat this procedure over a large number of samples of size 20, the true unknown value of p would be contained in 90% of the intervals; hence, we are confident that our particular interval, (0.175, 0.505), contains the true value of p. Note that we never use the term probability, since this interpretation is actually a frequentist one. The evaluation is based on samples that could have been observed but were not. Note also that since p is not a random quantity in the frequentist environment, p belongs (or not) to the interval without any probability attached to the statement. This is the reason to speak about confidence, not probability. Alternatively, to build Bayesian credible intervals (2, 3), consider a uniform prior in (0;1). This corresponds to normalizing the likelihood function, producing a beta posterior with, for example, a1⁄4 7 and b1⁄4 15. The 90% credible interval, the smallest interval with a posterior probability of 0.9, is (0.165, 0.483). This interval contains pwith a posterior probability of 0.9. In fact, this is the smallest interval that has 90%of the area under the likelihood function. This interval is a bit narrower than the confidence interval presented previously. We have shown that if one decides to use probabilities to replace confidences, the construction of the intervals is completely different than the usual method. In other words, Shrier et al.’s interpretation of the 95% confidence interval given on page 1204 of their article (1) and in their Appendix was technically incorrect. Rather than providing 95% confidence that the true value of the population parameter lies within the interval, the correct interpretation is that with the performance of equivalent studies, 95%of the observed confidence intervals would cover the true value of the parameter—a subtle but important difference, since population parameters are not random quantities and therefore probability statements should not be attached to them. Only in the Bayesian framework, which was not considered by Shrier et al., are parameters treated as random variables.

Independent Effect of Depression and Anxiety on Chronic Obstructive Pulmonary Disease Exacerbations and Hospitalizations

RATIONALE Depression and anxiety are significant comorbid and potentially modifiable conditions in chronic obstructive pulmonary disease (COPD), but their effects on exacerbations are not clear. OBJECTIVES To investigate the independent effect of depression and anxiety on the risk of COPD exacerbations and hospitalizations. METHODS A multicenter prospective cohort study in 491 patients with stable COPD in China. Multivariate Poisson and linear regression analyses were used, respectively, to estimate adjusted incidence rate ratios (IRRs) and adjusted effects on duration of events. MEASUREMENTS AND MAIN RESULTS Depression and anxiety were measured using the Hospital Anxiety and Depression Scale (HADS) at baseline. Other measurements included sociodemographic, clinical, psychosocial, and treatment characteristics. Patients were then monitored monthly for 12 months to document the occurrence and characteristics of COPD exacerbations and hospitalizations. Exacerbation was determined using both symptom-based (worsening of > or =1 key symptom) and event-based definitions (> or =1 symptom worsening plus > or =1 change in regular medications). A total of 876 symptom-based and 450 event-based exacerbations were recorded, among which 183 led to hospitalization. Probable depression (HADS depression score > or = 11) was associated with an increased risk of symptom-based exacerbations (adjusted IRR, 1.51; 95% confidence interval [CI], 1.01-2.24), event-based exacerbations (adjusted IRR, 1.56; 95% CI, 1.02-2.40), and hospitalization (adjusted IRR, 1.72; 95% CI, 1.04-2.85) compared with nondepression (score < or = 7). The duration of event-based exacerbations was 1.92 (1.04-3.54) times longer for patients with probable anxiety (HADS anxiety score > or = 11) than those with no anxiety (score < or = 7). CONCLUSIONS This study suggests a possible causal effect of depression on COPD exacerbations and hospitalizations. Further studies are warranted to confirm this finding and to test the effectiveness of antidepressants and psychotherapies on reducing exacerbations and improving health resource utilizations.

Delirium in older emergency department patients discharged home: effect on survival

OBJECTIVES: To determine whether prevalent delirium is an independent predictor of mortality in older patients seen in emergency departments (EDs) and discharged home without admission.

How big is too big? The perinatal consequences of fetal macrosomia.

OBJECTIVE The objective of the study was to examine the birthweight at which risks of perinatal death, neonatal morbidity, and cesarean delivery begin to rise and the causes and timing (antenatal, early or late neonatal, or postneonatal) of these risks. STUDY DESIGN This was a cohort study based on 1999-2001 US-linked stillbirth, live birth, and infant death records. Singletons weighing 2500 g or larger born to white non-Hispanic mothers at 37-44 weeks of gestation were selected (n = 5,983,409). RESULTS Infants with birthweights from 4000 to 4499 g were not at increased risk of mortality or morbidity vs those at 3500-3999 g, whereas those 4500-4999 g had significantly increased risks of stillbirth, neonatal mortality (especially because of birth asphyxia), birth injury, neonatal asphyxia, meconium aspiration, and cesarean delivery. Births at 5000 g or larger had even higher risks, including risk of sudden infant death syndrome. CONCLUSION Birthweight greater than 4500 g, and especially greater than 5000 g, is associated with increased risks of perinatal and infant mortality and morbidity.