Estimating effects of self-harm treatment from observational data in England: the use of propensity scores to estimate associations between clinical management in general hospitals and patient outcomes

Student thesis: Phd


Background: The use of health data from sources such as administrative and medical records to examine efficacy of health interventions is becoming increasingly common. Addressing selection bias inherent in these data is important; treatments are allocated according to clinical need and resource availability rather than delivered under experimental conditions. Propensity score (PS) methods are widely used to address selection bias due to observed confounding. This project used PS methods with observational cohort data relating to individuals who had attended an Emergency Department (ED) following self-harm (including self-poisoning and self-injury). This group is at greatly increased risks of further self-harm, suicide and all-cause mortality compared to the general population. However, it is not clear how hospital management affects risks of these adverse outcomes. Methods: A systematic review of PS methods with record-based mental health care data was used to determine the most appropriate methodological approach to estimate treatment effects following presentation to ED following self-harm. Following this review, PS stratification and PS matching methods were used with observational self-harm data to address observed baseline differences between patients receiving different types of clinical management following their hospital presentation (specialist psychosocial assessment, medical admission, referral to outpatient mental health services and psychiatric admission). Effects on repeat attendance for self-harm, suicide and all-cause mortality within 12 months were estimated. Advice on the interpretation and dissemination of results was sought from service users. Results: The systematic review resulted in 32 studies. The quality of the implementation and reporting of methods was mixed. Sensitivity analysis of the potential impacts of unobserved confounding was largely absent from the studies. Results from analysis of the self-harm cohorts showed that, broadly, prior to PS adjustment, individuals receiving each of the four categories of hospital management had higher risks of repeat attendance for self-harm, suicide and all-cause mortality than those not receiving that management. The use of PS methods resulted in attenuation of most of these increased risks. Psychosocial assessment appeared to be associated with reduced risk of repeat attendance for self-harm (risk ratio 0.87, 95% CI 0.80 to 0.95). Three advisors attended a group meeting and a further two provided responses by email. As a result of advisors' recommendations, an information sheet is being developed containing information about what patients can expect when attending hospital following self-harm and how treatment might influence future risk.Conclusions: Propensity score methods are a promising development in evaluating routine care for individuals who have self-harmed. There is now more robust evidence that specialist psychosocial assessment is beneficial in reducing risk of further attendances for self-harm. Advisors offered different perspectives to the researchers, leading to novel suggestions for dissemination.
Date of Award1 Aug 2017
Original languageEnglish
Awarding Institution
  • The University of Manchester
SupervisorNavneet Kapur (Supervisor), Richard Emsley (Supervisor) & Matthew Carr (Supervisor)


  • self-harm
  • suicide
  • psychosocial assessment

Cite this