Historical cohort study of trafficked patients in contact with secondary mental health services.
The study used data from the South London and Maudsley NHS Foundation Trust (SLaM) Biomedical Research Centre Case Register Interactive Search (CRIS) database . SLaM provides secondary mental health services to the London boroughs of Croydon, Lambeth, Lewisham and Southwark (a catchment area of approximately 1.2 million people), and has a near 100 % monopoly on provision. The CRIS database allows the searching and retrieval of anonymised patient records for over 200,000 patients in contact with SLaM services.
The study included SLaM service users whose clinical records indicated that they may have been trafficked for exploitation and who had one or more contact with SLaM services between 2007 and 2012. Trafficking was defined in accordance with the United Nations (UN) Optional Protocol to Prevent, Suppress and Punish Trafficking in Persons, Especially Women and Children (i.e. the recruitment or movement of people, by means such as force, fraud, coercion, deception, and abuse of vulnerability, for the purposes of exploitation), and included international and internal trafficking . Trafficking search terms (see Supplementary Information) were used to search the free-text clinical notes and correspondence of all patients in contact with SLaM services during the study period and to retrieve the records of patients whose records included one or more of the search terms. One researcher assessed the returned records for eligibility in the study (records that documented concerns that the patient may have been trafficked as per the UN definition of human trafficking); a second researcher (SO) independently assessed the eligibility of 10 % of the records. There were three scenarios by which healthcare professionals became aware that their patient had been trafficked: (1) the patient disclosed their experiences of exploitation; (2) the patient presented with signs of abuse or exploitation that led the professional to suspect trafficking, or (3) the healthcare worker was informed by another professional (e.g. law enforcement, immigration, social services, voluntary sector, other health professionals) that their patient had been trafficked. Less detail regarding the type of exploitation was typically recorded in the third situation, but correspondence between professionals included other relevant information that indicated that the patient met the study criteria e.g. that the patient was involved in criminal proceedings against their trafficker, was claiming asylum in relation to their experiences while trafficked, or was receiving social services or voluntary sector support as a victim of trafficking.
Data extraction and costing
Data were extracted on routinely recorded socio-demographic characteristics (e.g. gender, age, country of origin), clinical characteristics (e.g. International Classification of Disease-10 (ICD-10) diagnosis), mental health service characteristics (see Oram et al for full details ), and mental health service use. Mental health service use data included information on the date and duration of each contact, the type of professional contacted, the type of contact (inpatient, outpatient, accident and emergency or indirect contacts) and whether or not the patient attended. Data were also extracted from free-text clinical notes on patients’ experiences of physical and sexual violence prior to and during trafficking, and type of exploitation. Patients whose notes did not refer to violence prior to or during exploitation were categorised as not having experienced these types of abuse. Type of exploitation was categorised as sexual exploitation, domestic servitude, labour exploitation, financial exploitation (trafficking for benefit fraud, for example), or unknown.
Total costs were calculated by multiplying each resource use item by an appropriate unit cost. All unit costs, in United Kingdom (UK) pound sterling, were for the financial year 2012–2013 and included national NHS reference costs for hospital contacts  and national average unit costs for community health services . No adjustments were made for inflation but costs were discounted to reflect time preferences. Costs were assumed to occur at the beginning of each year , and the discount rate used was 3.5 %, based on the recommendations of the UK Treasury for the discounting of costs  .
Indirect contacts (phone calls, letters, faxes and emails) were not costed as the cost of these contacts are included in the published unit costs through the use of appropriate direct to indirect contact ratios. The cost of appointments not attended were assumed to be equal to the full cost of the appointment, which assumes the professional involved failed to make productive use of the time. This assumption was reduced in sensitivity analysis to zero but this had little impact on the results presented, so only the main analyses are presented.
Total costs over the period that each patient was in contact with SLaM services are presented as mean, standard deviation, median and range. Factors associated with total costs were explored using regression analysis. A list of possible cost predictors was created based on previous research investigating risk of mental health problems among trafficked people and in collaboration with clinical members of the research team [5, 15, 16]. This included: gender, age at first contact, diagnosis (psychotic disorder versus other), type of exploitation (sexual versus other [domestic servitude, financial exploitation, labour exploitation or unknown]), and violence pre- and during trafficking (sexual versus other). First, univariate associations between each of the specified predictors and total costs were explored in a linear regression. All variables are categorical except age at first contact, which is presented in two groups split at the median. In addition, age is presented split at the legal age for adulthood (<18 versus 18 and older) to assess any differences in the two populations. Secondly, multiple regression was used to reduce the variable set to those factors independently associated with mental health service costs. The model initially included all variables that had univariate associations with total costs at a significance level of 10 %, discarding from the model all variables that were no longer found to be important. Variables that did not have a univariate association were then added, one at the time, and retained if they added significantly to the model, otherwise discarded. The model derived was checked to ensure that no variables excluded would make a significant additional contribution . To confirm the validity of this approach, multiple regression was used with all independent variables included.
Cost data are commonly skewed and as a result the choice of regression method is not straightforward. Although the ordinary least squares assumptions may be violated, namely linearity and homoscedasticity, it is not appropriate to transform costs as analysis is then not concerned with the arithmetic mean but with the geometric mean, which is of less value to decision makers . For this reason, the results of the model were checked against the results obtained from a generalised linear model using an identity link function to describe the scale on which covariates in the model are related to costs and assuming a gamma distribution function for the costs . Results were compared with the results from a non-parametric bootstrap regression in order to assess the robustness of the confidence intervals and p-values to non-normality of the cost distribution.
Ethics and consent
Ethical approval for the research use of CRIS-derived anonymised databases without the written informed consent of SLaM service users was granted by an independent Research Ethics Committee (Oxfordshire C, reference 08/H0606/71). An Oversight Committee reviews all applications to use CRIS, and gave approval for this study (11/025).