Multiple imputation of incomplete multilevel data using Heckman selection models

Missing data is a common problem in medical research, and is commonly addressed using multiple imputation. Although traditional imputation methods allow for valid statistical inference when data are missing at random (MAR), their implementation is problematic when the presence of missingness depends on unobserved variables, that is, the data are missing not at random (MNAR). Unfortunately, this MNAR situation is rather common, in observational studies, registries and other sources of real-world data. While several imputation methods have been proposed for addressing individual studies when data are MNAR, their application and validity in large datasets with multilevel structure remains unclear. We therefore explored the consequence of MNAR data in hierarchical data in-depth, and proposed a novel multilevel imputation method for common missing patterns in clustered datasets. This method is based on the principles of Heckman selection models and adopts a two-stage meta-analysis approach to impute binary and continuous variables that may be outcomes or predictors and that are systematically or sporadically missing. After evaluating the proposed imputation model in simulated scenarios, we illustrate it use in a cross-sectional community survey to estimate the prevalence of malaria parasitemia in children aged 2-10 years in five regions in Uganda.

Arthroscopic partial meniscectomy vs non-surgical or sham treatment in patients with MRI-confirmed degenerative meniscus tears: a systematic review and meta-analysis with individual participant data from 605 randomised patients

OBJECTIVE

To identify subgroups of patients with magnetic resonance imaging (MRI)-confirmed degenerative meniscus tears who may benefit from arthroscopic partial meniscectomy (APM) in comparison with non-surgical or sham treatment.

METHODS

Individual participant data (IPD) from four RCTs were pooled (605 patients, mean age: 55 (SD: 7.5), 52.4% female) as to investigate the effectiveness of APM in patients with MRI-confirmed degenerative meniscus tears compared to non-surgical or sham treatment. Primary outcomes were knee pain, overall knee function, and health-related quality of life, at 24 months follow-up (0-100). The IPD were analysed in a one- and two-stage meta-analyses. Identification of potential subgroups was performed by testing interaction effects of predefined patient characteristics (e.g., age, gender, mechanical symptoms) and APM for each outcome. Additionally, generalized linear mixed-model trees were used for subgroup detection.

RESULTS

The APM group showed a small improvement over the non-surgical or sham group on knee pain at 24 months follow-up (2.5 points (95% CI: 0.8-4.2) and 2.2 points (95% CI: 0.9-3.6), one- and two-stage analysis, respectively). Overall knee function and health-related quality of life did not differ between the two groups. Across all outcomes, no relevant subgroup of patients who benefitted from APM was detected. The generalized linear mixed-model trees did also not identify a subgroup.

CONCLUSIONS

No relevant subgroup of patients was identified that benefitted from APM compared to non-surgical or sham treatment. Since we were not able to identify any subgroup that benefitted from APM, we recommend a restrained policy regarding meniscectomy in patients with degenerative meniscus tears.

How often can meta-analyses of individual-level data individualize treatment? A meta-epidemiologic study

BACKGROUND

One of the claimed main advantages of individual participant data meta-analysis (IPDMA) is that it allows assessment of subgroup effects based on individual-level participant characteristics, and eventually stratified medicine. In this study, we evaluated the conduct and results of subgroup analyses in IPDMA.

METHODS

We searched PubMed, EMBASE and the Cochrane Library from inception to 31 December 2014. We included papers if they described an IPDMA based on randomized clinical trials that investigated a therapeutic intervention on human subjects and in which the meta-analysis was preceded by a systematic literature search. We extracted data items related to subgroup analysis and subgroup differences (subgroup-treatment interaction p < 0.05).

RESULTS

Overall, 327 IPDMAs were eligible. A statistically significant subgroup-treatment interaction for the primary outcome was reported in 102 (36.6%) of 279 IPDMAs that reported at least one subgroup analysis. This corresponded to 187 different statistically significant subgroup-treatment interactions: 124 for an individual-level subgrouping variable (in 76 IPDMAs) and 63 for a group-level subgrouping variable (in 36 IPDMAs). Of the 187, only 7 (3.7%; 6 individual and 1 group-level subgrouping variables) had a large difference between strata (standardized effect difference d  ≥  0.8). Among the 124 individual-level statistically significant subgroup differences, the IPDMA authors claimed that 42 (in 21 IPDMAs) should lead to treating the subgroups differently. None of these 42 had d  ≥  0.8.

CONCLUSIONS

Availability of individual-level data provides statistically significant interactions for relative treatment effects in about a third of IPDMAs. A modest number of these interactions may offer opportunities for stratified medicine decisions.

Arthroscopic meniscectomy versus non-surgical or sham treatment in patients with MRI confirmed degenerative meniscus lesions: a protocol for an individual participant data meta-analysis

INTRODUCTION

Arthroscopic partial meniscectomy (APM) after degenerative meniscus tears is one of the most frequently performed surgeries in orthopaedics. Although several randomised controlled trials (RCTs) have been published that showed no clear benefit compared with sham treatment or non-surgical treatment, the incidence of APM remains high. The common perception by most orthopaedic surgeons is that there are subgroups of patients that do need APM to improve, and they argue that each study sample of the existing trials is not representative for the day-to-day patients in the clinic. Therefore, the objective of this individual participant data meta-analysis (IPDMA) is to assess whether there are subgroups of patients with degenerative meniscus lesions who benefit from APM in comparison with non-surgical or sham treatment.

METHODS AND ANALYSIS

An existing systematic review will be updated to identify all RCTs worldwide that evaluated APM compared with sham treatment or non-surgical treatment in patients with knee symptoms and degenerative meniscus tears. Time and effort will be spent in contacting principal investigators of the original trials and encourage them to collaborate in this project by sharing their trial data. All individual participant data will be validated for missing data, internal data consistency, randomisation integrity and censoring patterns. After validation, all datasets will be combined and analysed using a one-staged and two-staged approach. The RCTs' characteristics will be used for the assessment of clinical homogeneity and generalisability of the findings. The most important outcome will be the difference between APM and control groups in knee pain, function and quality of life 2 years after the intervention. Other outcomes of interest will include the difference in adverse events and mental health.

ETHICS AND DISSEMINATION

All trial data will be anonymised before it is shared with the authors. The data will be encrypted and stored on a secure server located in the Netherlands. No major ethical concerns remain. This IPDMA will provide the evidence base to update and tailor diagnostic and treatment protocols as well as (international) guidelines for patients for whom orthopaedic surgeons consider APM. The results will be submitted for publication in a peer-reviewed journal.

PROSPERO REGISTRATION NUMBER

CRD42017067240.