Instrumental variables II: instrumental variable application-in 25 variations, the physician prescribing preference generally was strong and reduced covariate imbalance

Brigham & Women's Hospital, Boston, MA 02120, USA.
Journal of clinical epidemiology (Impact Factor: 5.48). 05/2009; 62(12):1233-41. DOI: 10.1016/j.jclinepi.2008.12.006
Source: PubMed

ABSTRACT An instrumental variable (IV) is an unconfounded proxy for a study exposure that can be used to estimate a causal effect in the presence of unmeasured confounding. To provide reliably consistent estimates of effect, IVs should be both valid and reasonably strong. Physician prescribing preference (PPP) is an IV that uses variation in doctors' prescribing to predict drug treatment. As reduction in covariate imbalance may suggest increased IV validity, we sought to examine the covariate balance and instrument strength in 25 formulations of the PPP IV in two cohort studies.
We applied the PPP IV to assess antipsychotic medication (APM) use and subsequent death among two cohorts of elderly patients. We varied the measurement of PPP, plus performed cohort restriction and stratification. We modeled risk differences with two-stage least square regression. First-stage partial r(2) values characterized the strength of the instrument. The Mahalanobis distance summarized balance across multiple covariates.
Partial r(2) ranged from 0.028 to 0.099. PPP generally alleviated imbalances in nonpsychiatry-related patient characteristics, and the overall imbalance was reduced by an average of 36% (+/-40%) over the two cohorts.
In our study setting, most of the 25 formulations of the PPP IV were strong IVs and resulted in a strong reduction of imbalance in many variations. The association between strength and imbalance was mixed.

  • [Show abstract] [Hide abstract]
    ABSTRACT: Objectives Instrumental variable (IV) analysis is promising for estimation of therapeutic effects from observational data as it can circumvent unmeasured confounding. However, even if IV assumptions hold, IV analyses will not necessarily provide an estimate closer to the true effect than conventional analyses as this depends on the estimates' bias and variance. We investigated how estimates from standard regression (ordinary least squares [OLS]) and IV (two-stage least squares) regression compare on mean squared error (MSE). Study Design We derived an equation for approximation of the threshold sample size, above which IV estimates have a smaller MSE than OLS estimates. Next, we performed simulations, varying sample size, instrument strength, and level of unmeasured confounding. IV assumptions were fulfilled by design. Results Although biased, OLS estimates were closer on average to the true effect than IV estimates at small sample sizes because of their smaller variance. The threshold sample size above which IV analysis outperforms OLS regression depends on instrument strength and strength of unmeasured confounding but will usually be large given the typical moderate instrument strength in medical research. Conclusion IV methods are of most value in large studies if considerable unmeasured confounding is likely and a strong and plausible instrument is available.
    Journal of Clinical Epidemiology 11/2014; 67(11). DOI:10.1016/j.jclinepi.2014.05.019 · 5.48 Impact Factor
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: Instrumental variable estimates of causal effects can be biased when using many instruments that are only weakly associated with the exposure. We describe several techniques to reduce this bias and estimate corrected standard errors. We present our findings using a simulation study and an empirical application. For the latter, we estimate the effect of height on lung function, using genetic variants as instruments for height. Our simulation study demonstrates that, using many weak individual variants, two-stage least squares (2SLS) is biased, whereas the limited information maximum likelihood (LIML) and the continuously updating estimator (CUE) are unbiased and have accurate rejection frequencies when standard errors are corrected for the presence of many weak instruments. Our illustrative empirical example uses data on 3631 children from England. We used 180 genetic variants as instruments and compared conventional ordinary least squares estimates with results for the 2SLS, LIML, and CUE instrumental variable estimators using the individual height variants. We further compare these with instrumental variable estimates using an unweighted or weighted allele score as single instruments. In conclusion, the allele scores and CUE gave consistent estimates of the causal effect. In our empirical example, estimates using the allele score were more efficient. CUE with corrected standard errors, however, provides a useful additional statistical tool in applications with many weak instruments. The CUE may be preferred over an allele score if the population weights for the allele score are unknown or when the causal effects of multiple risk factors are estimated jointly. © 2014 The Authors. Statistics in Medicine published by John Wiley & Sons Ltd.
    Statistics in Medicine 11/2014; 34(3). DOI:10.1002/sim.6358 · 2.04 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: Abstract Objective: The purpose of this study was to assess the risk of manic switch associated with antidepressants in Medicaid-enrolled pediatric patients with bipolar depression. Methods: This retrospective cohort study involved 2003-2007 Medicaid Analytic eXtract (MAX) data from four geographically diverse states. The study sample included children and adolescents (ages 6-18 years) who had received a diagnosis of bipolar disorder on two or more separate occasions or during a hospital discharge, followed by a diagnosis of depression. According to the pharmacotherapy received by these patients in the 30 days around the index bipolar depression diagnosis, patients were categorized into five mutually exclusive groups. Manic switch was defined as having received a diagnosis of mania within 6 weeks after the initiation of bipolar depression treatment. Relative risks of manic switch between antidepressant monotherapy/polytherapy and their alternatives were assessed using Cox proportional hazards model. The robustness of the conventional Cox proportional hazards model toward possible bias caused by unobserved confounders was tested using instrumental variable analysis, and the uncertainty regarding manic switch definition was tested by altering the duration of follow-up. Results: After applying all the selection criteria, 179 antidepressant monotherapy, 1047 second-generation antipsychotic (SGA) monotherapy, 570 mood stabilizer monotherapy, 445 antidepressant polytherapy, and 1906 SGA-mood stabilizer polytherapy users were identified. In Cox proportional hazard analyses, both antidepressant monotherapy and polytherapy exhibited higher risk of manic switch than their alternatives (antidepressant monotherapy vs. SGA monotherapy, hazard ratio [HR]=2.87 [95% CI: 1.10-7.49]; antidepressant monotherapy vs. mood stabilizer monotherapy, HR=1.41 [95% CI: 0.52-3.80); antidepressant polytherapy vs. SGA-mood stabilizer polytherapy, HR=1.61 [95% CI: 0.90-2.89]). However, only the comparison between antidepressant monotherapy and SGA monotherapy was statistically significant. The instrumental variable analysis did not detect endogeneity of the treatment variables. Extending the follow-up period from 6 weeks to 8 and 12 weeks generated findings consistent with the main analysis. Conclusions: Study findings indicated a higher risk of manic switch associated with antidepressant monotherapy than with SGA monotherapy in pediatric patients with bipolar depression. The finding supported the clinical practice of cautious prescribing of antidepressants for brief periods.
    60th Meeting of American Academy of Child and Adolescent Psychiatry; 10/2013


1 Download
Available from