Practical considerations for sensitivity analysis after multiple imputation applied to epidemiological studies with incomplete data

Abstract

BackgroundMultiple Imputation as usually implemented assumes that data are Missing At Random (MAR), meaning that the underlying missing data mechanism, given the observed data, is independent of the unobserved data. To explore the sensitivity of the inferences to departures from the MAR assumption, we applied the method proposed by Carpenter et al. (2007).This approach aims to approximate inferences under a Missing Not At random (MNAR) mechanism by reweighting estimates obtained after multiple imputation where the weights depend on the assumed degree of departure from the MAR assumption.MethodsThe method is illustrated with epidemiological data from a surveillance system of hepatitis C virus (HCV) infection in France during the 2001–2007 period. The subpopulation studied included 4343 HCV infected patients who reported drug use. Risk factors for severe liver disease were assessed. After performing complete-case and multiple imputation analyses, we applied the sensitivity analysis to 3 risk factors of severe liver disease: past excessive alcohol consumption, HIV co-infection and infection with HCV genotype 3.ResultsIn these data, the association between severe liver disease and HIV was underestimated, if given the observed data the chance of observing HIV status is high when this is positive. Inference for two other risk factors were robust to plausible local departures from the MAR assumption.ConclusionsWe have demonstrated the practical utility of, and advocate, a pragmatic widely applicable approach to exploring plausible departures from the MAR assumption post multiple imputation. We have developed guidelines for applying this approach to epidemiological studies.

Cancer Risk for Fingolimod, Natalizumab, and Rituximab in Multiple Sclerosis Patients

“…We also tested the sensitivity to the ever‐treated exposure window by using an on‐treatment approach to follow‐up (patients were considered exposed between therapy start and 3 months after therapy stop) and, in the same analysis, restricting the therapy starts to the first of either of the therapies (ie, no previous exposure to the other therapies were allowed) in an attempt to capture only the effects of the active treatment. In post hoc analyses, we further investigated the sensitivity to large weights generated by the IPTW by limiting the maximum value of the stabilized weights to the 99th percentile of their original values and the sensitivity to the missing‐at‐random assumption of the multiple imputation, using the weighting method described by Carpenter et al with delta values chosen according to Héraud‐Bousquet et al…”

Characterizing and Managing Missing Structured Data in Electronic Health Records

“…The former case also meets the assumptions of most imputation methods but tends to be rare in practice. If a variable is MNAR, it may still be possible to impute, but the mechanism of missingness should be explicitly modeled and a sensitivity analysis is recommended to assess how much impact this could have on the final results [21,22]. While a statistical model of the mechanism of missingness is useful, there is no substitute for a deep familiarity with data at hand and how it was generated.…”

P2‐089: Impact of Home Visit Capacity on Genetic Association Studies of Late‐Onset Alzheimer’s Disease

“…When data are MNAR, deriving unbiased effect estimates relies on incorporating corrections for the missingness mechanism (i.e., what causes people to select out of clinic-based studies). Since the missingness mechanism is almost never known, this relies on sensitivity analyses exploring a range of possible missingness mechanisms [20]. Our results can help guide such sensitivity analyses in data sets that include only clinic-based data.…”