Objective: To evaluate the impact of sharing electronic health records (EHRs) with patients and map it across six domains of quality of care (ie, patient-centredness, effectiveness, efficiency, timeliness, equity and safety).

Design: Systematic review and meta-analysis.

Data sources: CINAHL, Cochrane, Embase, HMIC, Medline/PubMed and PsycINFO, from 1997 to 2017.

Eligibility criteria: Randomised trials focusing on adult subjects, testing an intervention consisting of sharing EHRs with patients, and with an outcome in one of the six domains of quality of care.

Data analysis: The Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines were followed. Title and abstract screening were performed by two pairs of investigators and assessed using the Cochrane Risk of Bias Tool. For each domain, a narrative synthesis of the results was performed, and significant differences in results between low risk and high/unclear risk of bias studies were tested (t-test, p<0.05). Continuous outcomes evaluated in four studies or more (glycated haemoglobin (HbA1c), systolic blood pressure (SBP) and diastolic blood pressure (DBP)) were pooled as weighted mean difference (WMD) using random effects meta-analysis. Sensitivity analyses were performed for low risk of bias studies, and long-term interventions only (lasting more than 12 months).

Results: Twenty studies were included (17 387 participants). The domain most frequently assessed was effectiveness (n=14), and the least were timeliness and equity (n=0). Inconsistent results were found for patient-centredness outcomes (ie, satisfaction, activation, self-efficacy, empowerment or health literacy), with 54.5% of the studies (n=6) demonstrating a beneficial effect. Meta-analyses showed a beneficial effect in effectiveness by reducing absolute values of HbA1c (unit: %; WMD=-0.316; 95% CI -0.540 to -0.093, p=0.005, I²=0%), which remained significant in the sensitivity analyses for low risk of bias studies (WMD= -0.405; 95% CI -0.711 to -0.099), and long-term interventions only (WMD=-0.272; 95% CI -0.482 to -0.062). A significant reduction of absolute values of SBP (unit: mm Hg) was found but lost in sensitivity analysis for studies with low risk of bias (WMD= -1.375; 95% CI -2.791 to 0.041). No significant effect was found for DBP (unit: mm Hg; WMD=-0.918; 95% CI -2.078 to 0.242, p=0.121, I²=0%). Concerning efficiency, most studies (80%, n=4) found either a reduction of healthcare usage or no change. A beneficial effect was observed in a range of safety outcomes (ie, general adherence, medication safety), but not in medication adherence. The proportion of studies reporting a beneficial effect did not differ between low risk and high/unclear risk studies, for the domains evaluated.

Discussion: Our analysis supports that sharing EHRs with patients is effective in reducing HbA1c levels, a major predictor of mortality in type 2 diabetes (mean decrease of -0.405, unit: %) and could improve patient safety. More studies are necessary to enhance meta-analytical power and assess the impact in other domains of care. PROTOCOL REGISTRATION: http://www.crd.york.ac.uk/PROSPERO (CRD42017070092).