Sengwee Toh

Risk of pre-eclampsia in first and subsequent pregnancies: prospective cohort study

Objective To investigate whether pre-eclampsia is more common in first pregnancies solely because fewer affected women, who presumably have a higher risk of recurrence, go on to have subsequent pregnancies. Design Prospective cohort study. Setting Swedish Medical Birth Register. Participants 763 795 primiparous mothers who had their first births in Sweden, 1987-2004. Main outcome measures Pre-eclampsia. Results The risk of pre-eclampsia was 4.1% in the first pregnancy and 1.7% in later pregnancies overall. However, the risk was 14.7% in the second pregnancy for women who had had pre-eclampsia in their first pregnancy and 31.9% for women who had had pre-eclampsia in the previous two pregnancies. The risk for multiparous women without a history of pre-eclampsia was around 1%. The incidence of pre-eclampsia associated with delivery before 34 weeks’ gestation was 0.42% in primiparous women, 0.11% in multiparous women without a history of pre-eclampsia, and 6.8% and 12.5% in women who had had one or two previous pregnancies affected, respectively. The proportion of women who went on to have a further pregnancy was 4-5% lower after having a pregnancy with any pre-eclampsia but over 10% lower if pre-eclampsia was associated with very preterm delivery. The estimated risk of pre-eclampsia in parous women did not change with standardisation for pregnancy rates. Conclusions Having pre-eclampsia in one pregnancy is a poor predictor of subsequent pregnancy but a strong predictor for recurrence of pre-eclampsia in future gestations. The lower overall risk of pre-eclampsia among parous women was not explained by fewer conceptions among women who had had pre-eclampsia in a previous gestation. Early onset pre-eclampsia might be associated with a reduced likelihood of a future pregnancy and with more recurrences than late onset pre-eclampsia when there are further pregnancies. Findings are consistent with the existence of two distinct conditions: a severe recurrent early onset type affected by chronic factors, genetic or environmental, and a milder sporadic form affected by transient factors.

The U.S. Food and Drug Administration's Mini‐Sentinel program: status and direction

The Mini‐Sentinel is a pilot program that is developing methods, tools, resources, policies, and procedures to facilitate the use of routinely collected electronic healthcare data to perform active surveillance of the safety of marketed medical products, including drugs, biologics, and medical devices. The U.S. Food and Drug Administration (FDA) initiated the program in 2009 as part of its Sentinel Initiative, in response to a Congressional mandate in the FDA Amendments Act of 2007.

Comparative Risk for Angioedema Associated With the Use of Drugs That Target the Renin-Angiotensin-Aldosterone System

BACKGROUND Although certain drugs that target the renin- angiotensin-aldosterone system are linked to an increased risk for angioedema, data on their absolute and comparative risks are limited. We assessed the risk for angioedema associated with the use of angiotensin converting enzyme inhibitors (ACEIs), angiotensin receptor blockers (ARBs), and the direct renin inhibitor aliskiren. METHODS We conducted a retrospective, observational, inception cohort study of patients 18 years or older from 17 health plans participating in the Mini-Sentinel program who had initiated the use of an ACEI (n = 1 845 138), an ARB (n = 467 313), aliskiren (n = 4867), or a β-blocker (n = 1 592 278) between January 1, 2001, and December 31, 2010. We calculated the cumulative incidence and incidence rate of angioedema during a maximal 365-day follow-up period. Using β-blockers as a reference and a propensity score approach, we estimated the hazard ratios of angioedema separately for ACEIs, ARBs, and aliskiren, adjusting for age, sex, history of allergic reactions, diabetes mellitus, heart failure, or ischemic heart disease, and the use of prescription nonsteroidal anti-inflammatory drugs. RESULTS A total of 4511 angioedema events (3301 for ACEIs, 288 for ARBs, 7 for aliskiren, and 915 for β-blockers) were observed during the follow-up period. The cumulative incidences per 1000 persons were 1.79 (95% CI, 1.73-1.85) cases for ACEIs, 0.62 (95% CI, 0.55-0.69) cases for ARBs, 1.44 (95% CI, 0.58-2.96) cases for aliskiren, and 0.58 (95% CI, 0.54-0.61) cases for β-blockers. The incidence rates per 1000 person-years were 4.38 (95% CI, 4.24-4.54) cases for ACEIs, 1.66 (95% CI, 1.47-1.86) cases for ARBs, 4.67 (95% CI, 1.88-9.63) cases for aliskiren, and 1.67 (95% CI, 1.56-1.78) cases for β-blockers. Compared with the use of β-blockers, the adjusted hazard ratios were 3.04 (95% CI, 2.81-3.27) for ACEIs, 1.16 (95% CI, 1.00-1.34) for ARBs, and 2.85 (95% CI, 1.34-6.04) for aliskiren. CONCLUSIONS Compared with β-blockers, ACEIs or aliskiren was associated with an approximately 3-fold higher risk for angioedema, although the number of exposed events for aliskiren was small. The risk for angioedema was lower with ARBs than with ACEIs or aliskiren.

Coronary Heart Disease in Postmenopausal Recipients of Estrogen Plus Progestin Therapy: Does the Increased Risk Ever Disappear?: A Randomized Trial

BACKGROUND Estrogen plus progestin therapy increases the risk for coronary heart disease (CHD) in postmenopausal women. However, this increased risk might be limited to the first years of use and to women who start therapy late in menopause. OBJECTIVE To estimate the effect of continuous estrogen plus progestin therapy on CHD risk over time and stratified by years since menopause. DESIGN Womens Health Initiative randomized, double-blinded, placebo-controlled trial. (ClinicalTrials.gov registration number: NCT00000611) SETTING 40 U.S. clinical centers. PATIENTS 16 608 postmenopausal women with an intact uterus at baseline from 1993 to 1998. INTERVENTION Conjugated equine estrogens, 0.625 mg/d, plus medroxyprogesterone acetate, 2.5 mg/d, or placebo. MEASUREMENTS Adherence-adjusted hazard ratios and CHD-free survival curves estimated through inverse probability weighting. RESULTS Compared with no use of hormone therapy, the hazard ratio for continuous use of estrogen plus progestin therapy was 2.36 (95% CI, 1.55 to 3.62) for the first 2 years and 1.69 (CI, 0.98 to 2.89) for the first 8 years. For women within 10 years after menopause, the hazard ratios were 1.29 (CI, 0.52 to 3.18) for the first 2 years and 0.64 (CI, 0.21 to 1.99) for the first 8 years, and the CHD-free survival curves for continuous use and no use of estrogen plus progestin crossed at about 6 years (CI, 2 years to 10 years). LIMITATION The analysis may not have fully adjusted for joint determinants of adherence and CHD risk. Sample sizes for some subgroup analyses were small. CONCLUSION No suggestion of a decreased risk for CHD was found within the first 2 years of estrogen plus progestin use, including in women who initiated therapy within 10 years after menopause. A possible cardioprotective effect in these women who initiated therapy closer to menopause became apparent only after 6 years of use. PRIMARY FUNDING SOURCE National Heart, Lung, and Blood Institute.

Selective serotonin reuptake inhibitor use and risk of gestational hypertension.

OBJECTIVE The purpose of this study was to assess the effects of treatment with selective serotonin reuptake inhibitors (SSRIs) on the risks of gestational hypertension and preeclampsia. METHOD The authors analyzed data from 5,731 women with nonmalformed infants and no underlying hypertension who participated in the Slone Epidemiology Center Birth Defects Study from 1998 to 2007. Gestational hypertension was defined as incident hypertension diagnosed after 20 weeks of pregnancy, with and without proteinuria (i.e., with and without preeclampsia). The risks of gestational hypertension and preeclampsia were compared between women who did and did not receive SSRI treatment during pregnancy. Relative risks and 95% confidence intervals (CIs) were estimated using the Cox proportional hazards model, adjusting for prepregnancy sociodemographic, lifestyle, reproductive, and medical factors. RESULTS Gestational hypertension was present in 9.0% of the 5,532 women who were not treated with SSRIs and 19.1% of the 199 women who were treated with SSRIs. Among women who received treatment, gestational hypertension was present in 13.1% of the 107 women who received treatment only during the first trimester and in 26.1% of the 92 women who continued treatment beyond the first trimester. The occurrence of preeclampsia was 2.4% among women who were not treated with SSRIs, 3.7% among women who were exposed to SSRIs only during the first trimester, and 15.2% among women who continued SSRI treatment beyond the first trimester. Relative to women who did not receive treatment, the adjusted relative risk of preeclampsia was 1.4 for women who discontinued treatment and 4.9 for women who continued treatment. CONCLUSION SSRI exposure during late pregnancy-whether a causal factor or not-might identify women who are at an increased risk for gestational hypertension and preeclampsia. Further investigation is needed in order to separate the effects of treatment with SSRIs from those of underlying mood disorders.

Causal Inference from Longitudinal Studies with Baseline Randomization

We describe analytic approaches for study designs that, like large simple trials, can be better characterized as longitudinal studies with baseline randomization than as either a pure randomized experiment or a purely observational study. We (i) discuss the intention-to-treat effect as an effect measure for randomized studies, (ii) provide a formal definition of causal effect for longitudinal studies, (iii) describe several methods -- based on inverse probability weighting and g-estimation -- to estimate such effect, (iv) present an application of these methods to a naturalistic trial of antipsychotics on symptom severity of schizophrenia, and (v) discuss the relative advantages and disadvantages of each method.

Antidepressant Use During Pregnancy and the Risk of Preterm Delivery and Fetal Growth Restriction

Objective: The associations between prenatal exposure to antidepressants and preterm delivery and fetal growth restriction are controversial and poorly understood. We studied the relation between antidepressant use and these outcomes. Methods: Analysis included women with nonmalformed infants interviewed in the Slone Epidemiology Center Birth Defects Study between 1998 and 2008. We estimated odds ratios (ORs) and 95% confidence intervals (CIs) for premature and small-for-gestational age (SGA) offsprings, adjusting for sociodemographic, lifestyle, medical, and reproductive factors. Results: The frequencies of preterm delivery were 7.3% among the 5710 nonusers (reference), 8.9% among the 192 selective serotonin reuptake inhibitor (SSRI) users (OR, 1.1; 95% CI, 0.6-2.0), and 15.3% among the 59 non-SSRI antidepressant users (OR, 2.2; 95% CI, 1.0-4.9); the respective frequencies of delivering an SGA offspring were 7.2%, 10.9% (OR, 1.7; 95% CI, 1.0-2.7), and 13.6% (OR, 2.2; 95% CI, 1.0-4.9). Compared with nonusers, the frequencies of preterm delivery (7.6%) and SGA offspring (5.7%) were not increased among the 106 women who discontinued SSRIs before the end of the first trimester. Among women who continued SSRIs beyond the first trimester, 10.5% delivered a preterm infant (OR, 1.3; 95% CI, 0.6-2.8) and 17.4% had an SGA offspring (OR, 3.0; 95% CI, 1.7-5.5). Conclusions: Women treated with SSRIs late in pregnancy had a higher frequency of delivering SGA infants, and women receiving non-SSRI antidepressants were more likely to deliver premature and SGA offsprings. The findings suggest an effect of underlying mood disorder or an effect common to both drug classes. In any case, prenatal antidepressant use may help identify women at elevated risks of delivering preterm and SGA infants.

Validity of health plan and birth certificate data for pregnancy research

To evaluate the validity of health plan and birth certificate data for pregnancy research.

Confounding adjustment via a semi-automated high-dimensional propensity score algorithm: an application to electronic medical records

A semi‐automated high‐dimensional propensity score (hd‐PS) algorithm has been proposed to adjust for confounding in claims databases. The feasibility of using this algorithm in other types of healthcare databases is unknown.

Potential Bias of Instrumental Variable Analyses for Observational Comparative Effectiveness Research

Key Summary Points Instrumental variable analysis is an increasingly popular method to establish causal conclusions from observational comparative effectiveness research (CER) studies. The instruments most commonly used in these studies are distance to facility, regional variation, facility variation, and physician variation. Instrumental variable analysis relies on several assumptions, some of which are empirically unverifiable and often suspect. The results of instrumental variable analyses may be biased substantially if the instrument and outcome are related through an unadjusted third variable: an instrumentoutcome confounder. Evidence of potential instrumentoutcome confounders was found for all 65 CER studies that used the 4 most common instruments and a mortality outcome. Findings from CER studies using instrumental variables should be evaluated critically for possible confounding. Patients, providers, and payers are increasingly relying on comparative effectiveness research (CER), which compares the benefits and risks of alternative clinical and health care delivery methods (1) to inform evidence-based health care decision making. Because CER is intended to improve patient care and guide health care resource allocation, its validity is crucial. Randomized, controlled trials (RCTs) are the gold standard for identifying the causal impact of a treatment or policy because random treatment assignment usually ensures that study and comparison groups are equivalent with respect to variables that affect the outcome (that is, confounders). However, because RCTs are not always feasible or generalizable, CER relies heavily on observational studies (2, 3), which are susceptible to confounding bias and other threats to validity (4). Instrumental variable analysis is recommended as a method to establish causal conclusions from observational CER studies (2, 57). As an example of this method, several studies have used relative distance to hospitals as an instrument in analyses aimed at estimating the effects on mortality of treatment with invasive cardiac procedures, specifically cardiac catheterization, after myocardial infarction (814). Researchers classify each hospital in the study region as a catheterization or noncatheterization hospital on the basis of the presence of a catheterization laboratory or the overall intensity (or volume) of catheterizations. Patients are assigned a value of the binary instrument based on whether they live closer to a catheterization hospital (making them more likely to receive the procedure) or a noncatheterization hospital. This instrumental variable analysis assumes that, similar to random assignment, the relative distance between a patients residence and a cardiac catheterization hospital predicts treatment choice independently of all characteristics (such as age, socioeconomic status, health status, or use of life-saving medications) that usually confound the relationship between treatment choice and outcome. In theory, instruments exploit variation in treatment assignment that allows causal inferences similar to those from RCTs. Random treatment assignment is an ideal instrument. In instrumental variable analysis done in observational settings, an instrument shares with experimental randomization certain characteristics that theoretically yield a causal inference. First, potential outcomes (15) for each patient (that is, the outcomes the patient would have under treatment and control conditions) are unrelated to the treatment status of other patients (the stable unit treatment value assumption). Second, the instrument affects receipt of the treatment of interest. Third, this effect is always in the same direction (monotonicity). Fourth, the instrument assigns treatment randomly, meaning that unobserved and observed patient characteristics that affect the outcome are similar in the treatment and comparison groups (ignorable treatment assignment). Finally, the instrument has an effect on the outcome only through the treatment assignment (the exclusion restriction) (16, 17). Although instrumental variable analysis is mathematically valid under these 5 assumptions (5, 1625), it is difficult to implement in practice (17, 26). There is a consensus that more research on the validity of instruments in observational CER is needed, particularly concerning violations of the ignorable treatment assignment and exclusion restriction assumptions (2, 5, 2729). At least one of these assumptions is violated if the instrument is related to the outcome through an unadjusted third variable, which we call an instrumentoutcome confounder (Figure 1). Instrumental variable analysis estimates of causal effects may be biased if an instrumentoutcome confounder has an effect on both the instrument and the outcome (violating the ignorable treatment assumption) or mediates an effect of the instrument on the outcome (violating the exclusion restriction). Although these assumptions are technically unverifiable, the identification of instrumentoutcome confounders through other sources provides evidence that an instrumental variable estimate may be biased. However, few researchers have searched for evidence of instrumentoutcome confounders outside their own, often limited, data (27, 29). Figure 1. Instrumental variable assumptions. The instrumental variable method substitutes actual random assignment to treatment with an instrument, a variable that predicts treatment assignment but is not related to all other factors that influence the outcome. This method relies on 5 critical assumptions (see text). Instrumental variable estimates of causal effects may be biased if a third variable, an instrumentoutcome confounder, has an effect on both the instrument and the outcome (violating the ignorable treatment assumption) or mediates an effect of the instrument on the outcome (violating the exclusion restriction). The instrumental variable analysis in the aforementioned example assumes that the association between relative distance to the hospital (the instrument) and mortality (the outcome) is due only to the effect of relative distance on treatment assignment after control for observed variables (Figure 1). A plausible instrumentoutcome confounder is rural residence. Patients living in rural areas are less likely to live close to a catheterization hospital; thus, the instrument is associated with rurality (8). Furthermore, ample evidence shows that rural residence is associated with several risk factors for mortality (3032). Therefore, an instrumental variable analysis would probably overstate the effect of catheterization because patients in the comparison group are, on average, sicker and more likely to die. In this study, we review relevant literature to identify instruments in CER, evaluate trends in the use of instruments in published CER studies, examine whether instrumental variable CER studies clearly state and attempt to address the assumption of no instrumentoutcome confounding, and identify the potential existence and effect of instrumentoutcome confounders for commonly used instruments. We list instrumentoutcome confounders that may compromise the validity of CER studies that use some of the most common instruments. We conclude by assessing the limitations and potential of instrumental variable analysis in CER. Methods Study Selection We conducted a systematic review in PubMed, EconLit, PsycINFO, Social Services Abstracts, Social Sciences Citation Index, and Web of Science to identify instrumental variable CER studies that were published in an English-language, peer-reviewed journal through 31 December 2011 and conducted in the United States and other industrialized countries. Specific search terms are provided in Table 1 of Supplement 1. Supplement 1. Systematic Review Search Strategies and Results We used the Institute of Medicines broad definition of CER, which includes both patient-level clinical interventions and system-level health care policies (1). We included noninterventional studies (for example, a study on the association between school junk food exposure and obesity) if the topic was amenable to clinical interventions or policy changes and the study included health-related outcomes. We excluded studies that were purely methodological, used only simulated data, or applied instrumental variable methods in an RCT. We also excluded studies that used Mendelian randomization as an instrument to elucidate biological mechanisms of disease (33) and studies that used an instrument to adjust for the effects of measurement error (34). Analysis of Instrumental Variable CER Studies We created a database of instrumental variable CER studies and catalogued them by year of publication, country, type of intervention, study population, type of instrument, strength of instrument, and outcome. We measured the trend in use of instruments in published CER studies by year and identified the most commonly used instrumentoutcome pairs. For each article that used one of these pairs, we also gathered information on the type of data set used and whether instrumental variable analysis was the sole type of analysis used. Finally, we assessed whether each instrumental variable CER article stated the assumption of no instrumentoutcome confounding and attempted to show, via additional analyses or discussion, whether the assumption was met. Identification of InstrumentOutcome Confounders Instrumentoutcome confounders are variables that are related to both the instrument and the outcome of interest, conditional on measured covariates. They violate the causal inference assumption that the instrument is independent of potential outcomes (15) and suggest that the instrument is not equivalent to random assignment (18). For the purposes of this paper, we included as instrumentoutcome confounders variables that have an effect on both the instrument and the outcome (such as rurality) or that mediate an effect of the instrument on the