Older forensic mental healthcare patients in England: demographics, physical health, mental wellbeing, cognitive ability and quality of life

Sampling and recruitment

A stratified cluster sampling frame was planned for community, low and medium secure units, taking into account gender mix and specialisation (e.g. patients with personality disorders or intellectual disabilities). This was disrupted due to COVID-19; however, we were able to recruit participants from a variety of settings across a geographically diverse range of sites. NHS Trusts were recruited through the Clinical Research Network (CRN). No specialist facilities were recruited, sites provided low, medium, and or high secure facilities and community care. Of the 12 community patients recruited, 11 were living independently, with one living in supported accommodation.

Local investigators liaised with members of the study team to identify patients aged over 55 years. These patients were then approached by local investigators to ascertain interest in participation, provide information sheets, and answer any questions about the study. Inclusion criteria for patients was those: aged 55 or over; under the care of forensic mental health services; able to complete self-report questionnaires and semi-structured interviews; who understood written and oral English; and who had capacity to consent. The CRN and PIs at different sites undertook recruitment and initial consenting for participation and as a result it was unknown how patients were approached and how many refused. The number of participants recruited from each trust is depicted in Table 1.

55 years was chosen as the cut-off as this reflects the expedited ageing experienced by forensic patients, suggested by some to be around 10 years (Merkt et al., 2020), which aligns with an often-used older age threshold in non-forensic populations used in research of 65 years. The ENHANCE study also included interviews with staff members from these services (reported separately) and patients known to these professionals were invited to take part. All participants were able to provide informed consent to complete questionnaires and to be interviewed. A sample size of N=36 was sought as this was considered sufficient to achieve saturated themes in the qualitative aspects of this project (reported elsewhere) and post-hoc power analyses report the obtained power of the correlations investigated in this study (see ‘Data analysis’).

Ethical approval

Ethical approval was granted by the NHS Health Research Authority (IRAS: 258016; REC: 19/EM/0350). Funding was provided by the National Institute for Health Research [PB-PG-1217-20028].

Data collection

Data collection took place between March 2020 and September 2021 across eight National Health Service (NHS) trusts. In total, 38 patients were recruited and completed all study questionnaires. However, it transpired one patient was 53 years old, so their data are excluded from the analysis. Meetings between the researcher responsible for data collection and participants took place in person (n=10), via video call (n=26), or over the phone (n=1). All patients gave informed consent, with written consent taken from those met face-to-face, and verbal recorded consent from those met via video call or phone. Both methods of recording consent were approved by the relevant ethics committee.

Clinical, legal and demographic data (Tomlin et al., 2022) were extracted from patient clinical records by principal investigators based at each recruitment site. Legal data included length of stay in the service, nature of the index offence(s) and Mental Health Act 1983 status. Index offences were categorised according to the UK Home Office Offence Classification Index. Where a patient had more than one index offence, we report the most severe as indicated by the Home Office Crime Severity Score. It should be noted that in England and Wales, patients do not need to have committed an index offence to receive treatment in forensic services. They might receive treatment in these services under a civil, non-forensic legal section where they are at risk of harm to themselves or others, which cannot be safely managed in general psychiatric settings.

Clinical data included ICD-10 diagnoses, body mass index (BMI), lists of physical health conditions and medication data (total number of drugs currently prescribed, number of psychotropic drugs currently prescribed, and Anticholinergic Effects on Cognition scores (Bishara et al., 2017)). Medications were included if prescribed for regular consumption. As required (pro re nata or prn) prescriptions were not counted as it would not be possible to ascertain how much of the drug had actually been administered.

Patients completed six questionnaires:

Due to disruptions due to COVID-19, our data collection methods had to be revised during data collection. Two sets of data were collected in person, where the participants completed the four self-report questionnaires (Q1 EQ-5D-5L; Q2 ReQol; Q3 SWEMWBS; Q4 FRQ), and the researcher undertook the two (Q5 MoCA; and Q6 CCRT) other questionnaires with the participants (Q5 and Q6 both require the researcher to actively administer them). The rest of the questionnaires were collected over video-call. For this, the researcher administered Q5 and Q6 to the participants, but Q1–Q4 were filled in by the participants independently to the call. Q1–Q4 were distributed to the participants by principal investigators on each site and were posted/emailed back to the research team once completed.

Data analysis

Post-hoc power analyses were conducted as the number of completed questionnaires varied (e.g., SWEMWBS n =36, and FRQ n =27). G*Power (Faul et al., 2007) suggested that at r = 0.5, p = 0.05, two-tailed, n = 36, our analysis yielded a sufficient power of 0.89. G*Power also indicated at r = 0.5, p = 0.05, two-tailed, n = 27, that analyses including the FRQ yielded a power of 0.78, just under the usually accepted standard of 0.8 (in this study achieved with a sample of n =28). This suggests that whilst most correlations conducted in this study are sufficiently powered, our findings involving the FRQ should be seen as exploratory.

Three percent of the questionnaire data were missing, largely due to nine patients not completing the Short CCRT, reporting difficulties with eyesight, reading level or not wanting to complete this. Eight percent of the demographic, legal and clinical data were missing due to recording issues in patient files. Pairwise deletion was used to handle missing data. IBM’s statistical package for the social sciences (SPSS) software v.27 was used. The distribution of questionnaire response data was assessed with the Shapiro-Wilk statistic and most variables were non-normally distributed. To account for this, non-parametric methods were used.

The internal consistency of the questionnaires used in the study was assessed prior to analysis. This suggested all questionnaires were appropriate to use: SWEMWBS, α= .886, n=36; EQ-5D-5L, α= .871, n=36; ReQoL, α= .859, n=36; FRQ, α= .945, n=27; and MoCA, α= .660, n=34. The alpha value (α) for the MoCA was lower than for the other measures and much of the literature (Nasreddine et al., 2005) but other studies (Bernstein et al., 2011) have reported values lower than α=.7 (the recommended cut-off for assuming adequate internal consistency; Bland & Altman, 1997). Spearman’s RHO (ρ) was used to assess correlations². Effect sizes (r) were judges as follows: 0.1, 0.3, and 0.5 as small, medium and large (Cohen, 1988). Statistical significance was set at p=<0.05; effect sizes are reported where appropriate. All correlation coefficients reported without a p value are significant at p=<0.001.

Descriptive statistics

Table 2 presents the demographic, clinical and legal profiles of the patient group. Table 3 and Table 4 give an overview of the mental health and physical diagnoses given to patients in the sample. These show that participants were mostly men (92%), of white British ethnicity (81%), with a mean age of 60 years. Median length of stay in current institution was 1404 days (approximately 45 months). The most frequently diagnosed mental disorders were schizophrenia, schizotypal, and delusional disorders (60%), any type of personality disorder (41%), and then mental and behavioural disorders due to psychoactive substance use and mood (affective) disorders (both at 16.2%)³. Scores for each questionnaire are presented in Table 5.

Mental wellbeing

Mental wellbeing (SWEMWBS) was significantly positively associated with recovery-related quality of life (ReQoL) (ρ= .773), and ‘health today’ (ρ= .486)⁴ as assessed by the EQ-5D-5L, both considered at or above large effect sizes. It was negatively correlated with the depression and anxiety domain (ρ= -.348) of the EQ-5D-5L and was trending towards a significant relationship with the ‘usual activities’ domain of the same measure (meaning fewer problems in these domains; ρ= -.324, p=.054). A negative correlation was also observed for inpatient perceptions of restrictiveness in care (ρ= -.481)), a large effect but the small sample size must be borne in mind here. The association between mental wellbeing and mild cognitive impairment was trending towards significance in a negative direction (ρ= -.320, p=.065).

Recovery-related quality of life

Recovery-related quality of life positively correlated with the EQ-5D-5L ‘health today’ domain (ρ= .627), an above large effect size, and its overall index value (ρ= .362). Recovery-related quality of life was negatively correlated with ‘mobility’ (ρ= -.340), ‘usual activities’ (ρ= -.551), ‘depression and anxiety’ domains on the EQ-5D-5L (ρ= -.408) implying fewer problems on these domains, and (for inpatients) experiences of restrictiveness (ρ= -.608), a large effect but the small sample size must be borne in mind here. The association between recovery-related quality of life and mild cognitive impairment was also significant in a negative direction (ρ= -.377).

Health status and perceptions of physical wellbeing

Nearly half (49%) our sample were diagnosed with diabetes, 38% had a disease of the cardiovascular system, and one-fifth (19%) had high cholesterol (e.g. hypercholesterolemia, hyperlipidaemia, raised triglycerides). Around 16% had COPD and 14% had some form of visual impairment. BMI data were available for 30 patients; the mean score for our sample was 31.7, classified as ‘obesity class one’ by the World Health Organisation (WHO). Nine patients met the threshold for ‘pre-obesity’, and 19 patients for obesity classes one, two or three. Only two patients were in the ‘normal weight’ range. On average, patients were prescribed 2.1 psychotropic medications for regular use, with an average anticholinergic effect on cognition score of 2.4, according to the scoring system described in (Bishara et al., 2017).

The EQ-5D-5L domains mobility, self-care, usual activities, pain and discomfort, anxiety and depression were all significantly positively correlated with each other. They were all also negatively correlated with the broader EQ-5D-5L indicators of ‘health today’ and the overall ‘index value’ (a measure of overall wellbeing). Patients' assessment of their health on that specific day was linked with cognitive impairment scores (ρ= -.432) whilst their overall health index score on the same questionnaire was not.

Demographic characteristics and patient outcomes

Age was not significantly linked to any of the outcomes measured. Length of stay in current setting was positively associated with usual activities (ρ= .440) and anxiety and depression (ρ= .384) indicating a greater number of problems in these domains, and negatively associated with EQ-5D-5L index value suggesting poorer health (ρ= -.374). Despite the correlation with the index value, there was no significant relationship with patients' assessments of their health on that specific day.

Mild cognitive impairment scores were negatively correlated with recovery-related quality of life scores (ρ= -.377) and positively with higher restrictiveness ratings (for inpatients, ρ=.474). Premorbid IQ was negatively correlated with the EQ-5D-5L ‘health today’ domain (ρ= -.432). As our sample was too small to conduct analyses of difference between more than two groups (i.e., ANOVA), we present median recovery-related quality of life, overall wellbeing (EQ-5D-5L index value), mental wellbeing, and experiences of restrictiveness scores across treatment settings and levels of leave in Table 6. Though it is not possible to draw firm conclusions from these findings, at face value there appears to be a trend indicating that scores improve as levels of security decrease from high to low and as levels of leave increase. The exception to this is that community patients appear to have equivalent or poorer outcomes on the SWEMWBS and EQ-5D-5L than inpatients, and ReQoL scores at a level between patients in medium and low security.

Clinical and research implications

To summarise the practical implications of our study, 65% of our sample had possible cognitive impairment according to a validated measure. This suggests that older patients might benefit from interventions to improve cognition or ameliorate cognitive decline, though more evidence is needed to speak to the efficacy of different interventions. Studies suggest improvement might be best achieved through cognition-based interventions, physical exercise and antioxidants (Xu et al., 2021). Patient recovery-related quality of life and mental wellbeing is likely enhanced by engagement in a range of meaningful and age- or needs-appropriate activities that include work, study, housework, family or leisure activities. Services should continue to address physical healthcare needs, especially relating to cardiovascular health, as patients progress into the community to ensure that physical health concerns do not hinder mental wellbeing and recovery.

The negative correlation of recovery-related quality of life with problems related to depression and anxiety, ability to perform usual activities (work, study, housework, family or leisure) and mobility highlights the importance of seeking to address these issues to enhance quality of life. To reduce levels of obesity and diabetes, more consideration should be given to improving patient physical activity levels, diet, and sleep quality. Acknowledging that many patients will be experiencing cognitive impairment, services should make allowances for this in provision of services, needs assessment, risk assessment, interventions, and treatment, as well as providing relevant staff training.

Further research should address participation in meaningful activities in more detail. Tools that measure aspects of engagement in occupational activities such as the Model of Human Occupation Screening Tool (MOHOST; see: Fan et al., 2016) or the Engagement in Meaningful Activity Survey (EMAS; see: O’ Flynn et al., 2018) should be used. These studies should be longitudinal and compare age groups. This would complement the growing literature assessing levels of met and unmet need in this population (Girardi et al., 2018). Other research could evaluate CST-based interventions, potentially delivered via iPad app, in this group.

Limitations

Our study has some limitations. Our sample size can be considered relatively small (N=37), precluding the use of multivariate analysis (e.g. regression) or comparisons of mean differences across multiple groups (e.g. ANOVA). It is possible that given the small sample size, factors not included in this study or not controlled for played a role in shaping patient experiences of for example, quality of life and wellbeing. Having acknowledged this, post-hoc power analyses indicate that most of the analyses included in the study attain the generally accepted power of 0.8 and our sample size is comparable to other studies of this population (Coid et al., 2002; Das et al., 2011; Das et al., 2012; Di Lorito et al., 2019; Girardi et al., 2018; Lightbody et al., 2010; Tomar et al., 2005), most of which are retrospective using hospital records and did not involve active participant recruitment. The internal consistency of the MoCA was questionable (α=.67), falling just below the generally accepted α=.70 (Bland & Altman, 1997). Thus, conclusions concerning mild cognitive impairment in this study should be read with some caution. Nonetheless, our findings in this regard are similar to other studies (Di Lorito et al., 2019). Finally, we did not compare our sample to a representative or whole population of all older patients in England and Wales or a representative or whole population sample of all patients; nor did we obtain a random sample of participants. This places limitations on the extent to which we can conclude our findings are generalisable to patients outside our sample.

Underlying data

Open Science Framework: Older adult forensic mental health patients: defining needs, barriers, facilitators and 'what works' to enable better quality of life, health and wellbeing and to reduce risk, https://doi.org/10.17605/OSF.IO/GS37Y. (Tomlin et al., 2022)

This project contains the following underlying data:

Extended data

Open Science Framework: Older adult forensic mental health patients: defining needs, barriers, facilitators and 'what works' to enable better quality of life, health and wellbeing and to reduce risk, https://doi.org/10.17605/OSF.IO/GS37Y. (Tomlin et al., 2022)

This project contains the following extended data:

Data are available under the terms of the Creative Commons Attribution 4.0 International license (CC-BY 4.0).