Background Publication bias has been demonstrated in several high-profile cases associated with clinical trials of pharmaceuticals, but empirical evidence on this bias is scarce in HSDR, which concerns research to produce evidence on the quality, accessibility and organisation of health services. We set out to gauge the existence and scale of publication bias in quantitative HSDR by following up the publication status of four cohorts of studies identified from registries and conferences.

Methods We identified HSDR cohorts from four sources: (1) all quantitative studies funded by the NIHR HS&DR Programme (and its predecessors) between 2007–2014 (n=100); (2) a random sample of quantitative HSDR studies registered with HSRProj (a US-based prospective registry of health service and public health research) and completed in 2012 (n=100); (3) abstracts from the 2012 International Society for Quality in Health Care (ISQua) conference (n=50) and (4) Health Services Research UK (HSRUK) conference, 2012–14 (n=50). We checked the publication status of selected studies in 2018 by searching PubMed and Google. When no relevant publications were found, we contacted the investigators to verify the publication status. We compiled descriptive statistics for each cohort and used univariate analysis to explore the association between reporting of statistically significant findings and publication status.

Results Following online search of publications, we needed to contact investigators for 145/300 (48%) of the studies and had a response rate of 60% (67/111) among those with valid contact details. Publication of findings in academic journals varies between cohorts: NIHR 64%, HSRProj 75%, ISQua 26% and HSRUK 70%. When grey literature (including technical reports available online) was taken into account, the publication rates were NIHR 100%, HSRProj 91%, ISQua 32% and HSRUK 80%. Statistically significant findings were reported by 79% (237/300) of the studies. There was no apparent association between reporting of statistically significant findings and publication in academic journals for the combined NIHR/HSRProj cohort (OR 1.19, 0.50–2.81) and conference cohort (OR 0.88, 0.25–3.10).

Discussion The rates of publication were generally high among HSDR studies that we sampled (which are likely to represent the high rectitude end of distribution), except in the ISQua cohort where the originators of the studies tended to be service providers rather than academic researchers. The power of our exploratory analysis was limited by the relatively small number of studies reporting non-significant findings and/or having a non-publication status. The publication policy of research funders (e.g. NIHR) appears to be effective in ensuring the publication of research findings.