paper from the National Confidential Inquiry into Suicide and Homicide (NCISH) published in The Lancet Psychiatry starts from the hypothesis that implementation of service changes associated with improvements in patient safety have led to a reduction in the rate of suicide. It manages to show an association of about 20-30% reduction in suicide between 1997-2012 with 16 policies and procedures that relate to ward safety (eg. removing non-collapsible curtain rails), availability of community services (eg. implementing a Crisis Resolution and Home Treatment team within community health services), staff training (eg. training clinical staff in suicide risk management), adoption of specific policies (eg. policy regarding response to inpatients who abscond), and adoption of The National Institute for Health and Care Excellence (NICE) guidelines (eg. NICE depression guidelines). An accompanying comment paper hails this finding as a success for clinical governance.
But what if this 20-30% reduction in suicides in the clinical population happened for other reasons rather than anything to do with these 16 service changes? The fact that the incidence rate ratio was very similar for all 16 policies could be said to support this inference of lack of causal effect. The article notes that the incidence rate ratios were higher for the general population than for the patient population in the study, but the patient population in mental health services has also changed over recent years, with probably more minor cases being referred.
Although the paper does acknowledge that this "study was observational therefore we cannot make causal inferences", I can't find any specific mention of the fact that suicide rates were falling during the period under study. If the same study had been done during a period of rising suicide rates, the service changes would have been associated with an increase rather than decrease in suicide. And by writing that "service delivery variables are associated with suicide rates", the paper, as does the comment paper, leads people to think that a causal connection is being inferred, which is what the original hypotheses was. But the study isn't a hypothesis testing paradigm, as the paper notes, because the use of "randomised controlled designs for this research would be extremely challenging".
NCISH seems to have a habit of using data to justify its own prejudices (see previous post). Whatever happened to the principle of scientific scepticism? And how does a paper like this get through The Lancet Psychiatry peer review process?