Estimating provisional causal treatment effects on reoffending for binary and multiple treatments using quasi-experimental methods for people in prison with an alcohol use disorder in England

Aim:
Little is known as to the efficacy of criminal justice treatment interventions for people in prison with an alcohol use disorder (AUD) to reduce reoffending. Prison-based treatment aims to provide pharmacological and psychosocial interventions that address physical and mental health needs concurrently with approaches to deal with recidivism. This thesis determines whether there is a ‘provisional’ causal treatment effect in reduced re-offending post-release for people in prison receiving treatment for an AUD. Quasi experimental methods that derive a treatment effect will be assessed for a single (or binary) treatment system and where multiple treatments exist. The validity of using these methods in criminal justice settings will also be assessed.

Method:
Two quasi-experimental methods were applied; (1) a quasi-experimental propensity score matched (PSM) observational study to calculate the average treatment effect on reoffending compared to a matched control group and (2) multiple treatment effect estimators for pharmacological treatment only; Risk-Need-Responsivity (RNR) compliant treatment; and ‘other’ psychosocial interventions compared to a control group. 26,654 people in prison with an AUD were linked across five management information systems with 56-60% successfully matched across databases. For PSM, a one-to-one match without replacement and a marginal Cox proportional hazards time-to-event model were used with only treatment as a prognostic and the inverse proportional weights (derived from a one-to-one matching) as covariates as recommended by Austin (2010). Four regression-based treatment effect estimators were deployed to compare the effectiveness of the three treatment groups compared to controls. The analysis was supplemented by qualitative interviews with operational and strategic stakeholders involved in prison-based AUD treatment.

Results:
Overall, no statistically significant difference in reoffending rates and risk of reoffending on release were noted. The findings were dependent on whether static or dynamic covariates were included. The outcome for RNR-compliant treatment suggest a lower recidivism rate compared to the control group. Pharmacological-only treatment results in a statistically significant higher level of reoffending relative to the untreated group. Regression-based methods of deriving a treatment effect from observational studies are potentially affected by sample size. The treatment effect using propensity score matching is affected by the number of covariates included in the model. Poor quality or missing data were shown to adversely affect confidence levels. The use of quasi-experimental methods to determine a treatment effect is subject to a wide range of researcher assumptions and requires transparency to ensure there is sufficient equivalence (balance) across treated and comparison groups.

Conclusion:
The creation of a universal system of ‘equivalence of care’ framed within a public health context in English prisons may have had an unintended consequence of diluting approaches that reduce recidivism. People in prison with an AUD present with different needs than other segments of substance misuser. There is an opportunity to develop an AUD-focused and cross-disciplinary model for prison-based treatment that unites public health and effective crime reduction approaches. Quasi-experimental methods provide an alternative to randomisation but require methodological enhancements to reliably calculate a treatment effect.