Including non-randomized studies in network meta-analyses: a case of sparsely reported adverse event data

Session: 

Oral session: Inclusion of non-randomized designs

Date: 

Sunday 16 September 2018 - 14:00 to 14:20

Location: 

Sidlaw Auditorium

All authors in correct order:

Watt J1, Goodarzi Z2, Veroniki A3, Tricco A1, Straus S1
1 Li Ka Shing Knowledge Institute, St. Michael’s Hospital, Canada
2 University of Calgary, Canada
3 University of Ioannina, Greece
Presenting author and contact person

Presenting author:

Jennifer Watt

Contact person:

Jennifer Watt
Abstract text
Background: Risk of harm from adverse drug events is often sparsely reported in randomized controlled trials (RCTs).

Objectives: Our objectives were to:
1) compare the results of network meta-analyses (NMAs) including, a) RCTs and b) RCTs and non-randomized studies (NRSs); and
2) demonstrate the importance of NRSs in NMAs of safety outcomes.

Methods: We conducted a systematic review of the literature from database inception to 8 January 2018, to retrieve studies assessing the safety of medications for the behavioral and psychological symptoms of dementia (BPSD) (PROSPERO: CRD42017050130). Our primary outcome was risk of fracture. Studies were retrieved from MEDLINE, Embase, PsycINFO, CINAHL, CCTR, CDSR, DARE, and the grey literature. We completed screening, data abstraction, and risk of bias assessments independently in duplicate. A Bayesian random-effects NMA model was used to derive posterior median odds ratios (OR) for the risk of fracture in RCTs. We used a three-level hierarchical shared parameter random-effects NMA model to derive ORs for the risk of fracture from both RCTs and NRSs. We used minimally-informative priors for all between-study and between-study type heterogeneity parameters in our NMAs. We compared the posterior mean deviances of the inconsistency and consistency models.

Results: We screened 17210 abstracts and retained 3918 for full-text review; 28 RCTs (12,336 participants including 136 with fractures) and six NRSs (28154 participant including 2788 with fractures) contributed 10 treatment nodes to our NMAs. The RCTs and NRSs were at moderate risk of bias. In the NMA of RCTs, atypical antipsychotics were associated with a lower odds of fracture compared to placebo (OR 0.34, 95% credible interval (CrI) 0.12 to 0.93) and cholinesterase inhibitors (OR 0.32, 95% CrI 0.10 to 0.95). However, in the NMA of both RCTs and NRSs, there was no significant risk of fracture associated with any treatments and we identified no overall inconsistency.

Conclusions: Based on combined RCT and NRS evidence, there was no increased risk of fracture from medications used to treat BPSD. Systematic reviewers should consider including NRSs in their NMAs, especially if their topic is an area with few RCTs and/or number of events.

Stakeholder involvement: Care partners of patients with dementia selected our most important safety outcomes.

Relevance to patients and consumers: 

Behavioral and psychological symptoms of dementia (BPSD) (e.g. anxiety, agitation, depression, apathy) occur in up to 75% of patients with dementia each month. Twelve care partners (e.g. nurse, caregiver, physician) involved in the care of patients with BPSD were surveyed to identify the most important harmful outcomes associated with medications used to treat BPSD. Risk of fracture was selected as the most important outcome for our study based on their input. We demonstrated there was no increased risk of fracture from any medication. This information will inform shared decision-making between patients, caregivers, and clinicians.