1. Name:Shanil Ebrahim, MSc, MSc, PhD Departments of Clinical Epidemiology & Biostatistics, and Anesthesia McMaster University, Hamilton, Canada Stanford Prevention Research Center, Department of Medicine Stanford University, Stanford, USA Authors:Shanil Ebrahim, Elie Akl, Reem Mustafa, Xin Sun, Stephen Walter, Diane Heels-Ansdell, Pablo Alonso-Coello, Bradley Johnston, Gordon Guyatt Disclosure:No conflicts of interest Addressing missing participant data in systematic reviews: Part II – Continuous outcomes

2. Missing participant data 87% of RCTs published in top general medical journals suffer from missing participant data (Akl et al, BMJ 2012) May introduce bias in trials and systematic reviews Overestimate in positive trials if: o intervention do badly, control do well Plausible worse case Approach for systematic reviews o Plausible worse case in each study, then pool estimates 33% may change

3. Continuous Missing Data Methods for investigating extent missing participant data for continuous outcomes in systematic reviews? Cochrane handbook: Goal: fill the void NONE NO GUIDANCE

4. Challenges Multiple imputation techniques not applicable Approach should ideally address: o Measure of effect o Associated measure of precision

5. Development of approach Consider sources for imputations Strategies to handle control and intervention separately Complete case analysis followed by progressively more stringent strategies to challenge estimates

6. Effect 5 sources of data reflecting real observed mean scores in participants followed-up in individual trials in a meta-analysis: Ranging from: o Best mean score among intervention arms o Worst mean score among control arms Precision Median SD (plausible) Imputing effect & precision

7. Developed 4 progressively more stringent imputation strategies for participants with missing data in both arms Imputation strategies Assumptions about the means of participants in CONTROL Assumptions about the means of participants in INTERVENTION C: Mean score from the control arm of the same trial D: Worst mean among intervention arms E: Worst mean among control arms A: Best mean among intervention arms B: Best mean among the control arms C: Mean score from the control arm of the same trial 1 Intervention and control : Mean score from the control arm of the same trial 2 Intervention : Worst mean among intervention arms Control : Best mean among control arms 3 Intervention : Worst mean among control arms Control : Best mean among control arms 4 Intervention : Worst mean among control arms Control : Best mean among intervention arms

8. Combining observed & imputed data 3-step method for each strategy: [1] Combine observed means and SDs of those with available data with imputed means and SDs for those with missing data [2] Use pooled estimates to calculate treatment effect per study [3] Perform a standard random-effects meta-analysis to pool

9. Application of approach: 1 Cognitive behavioural therapy (CBT) versus minimal or no treatment for depression in patients receiving disability benefits 8 RCTs: Beck Depression Inventory Median missing participant data rate = 21% (range 0 to 41%)

15. Finasteride therapy versus placebo on improvement in scalp hair for men with androgenetic alopecia 8 RCTs Median missing participant data rate = 14% (range 0% to 24%) Application of approach: 2

17. CBT review: o Effect diminished, lost significance as strategies became more stringent o Rate down for risk of bias Finasteride review: o Even most stringent: statistical sig remains, effect important o Do not need to rate down for risk of bias Discussion

19. Extension of approach Limitation to reviews using one instrument to assess outcome Challenges Imputations for both effects and precision Pooling effects from different instruments Approach Convert scores to units of a reference measurement instrument (Thorlund et al, Research Synthesis Methods, 2012) Apply 4 increasingly stringent imputation strategies Also includes an approach to consider the MID to determine the proportion who have an important benefit

20. Extension of approach:

21. shanil.ebrahim@mcmaster.ca Hands-on Exercise