Introduction
Common mental health disorders (CMDs) include anxiety and depressive and panic disorders, and affect at least one in eight adults globally (Institute of Health Metrics and Evaluation, 2019). In England, the NHS launched the Improving Access to Psychological Therapies (IAPT) programme in 2007, following a report outlining the importance of mental health on a societal and economic level (Layard, Reference Layard2005). IAPT services provide evidence-based psychological therapies recommended by the National Institute of Clinical Excellence (NICE) (Clark, Reference Clark2011). Psychological therapies, such as cognitive behavioural therapy (CBT) and behavioural activation, have proven short- and long-term positive impacts on CMD sufferers, yielding outcomes better than or equal to medication alone (Dwight-Johnson et al., Reference Dwight-Johnson, Sherbourne, Liao and Wells2000; Wiles et al., Reference Wiles, Thomas, Abel, Ridgway, Turner, Campbell and Lewis2013) and with positive economic outcomes (Layard, Reference Layard2006). IAPT services have now treated over one million patients and have published results showing excellent outcomes (Clark et al., Reference Clark, Layard, Smithies, Richards, Suckling and Wright2009; Gyani et al., Reference Gyani, Shafran, Layard and Clark2013). The programme continuously seeks to improve those outcomes. Understanding what factors influence outcomes is important in understanding how clinical care could be improved at the patient level, in understanding variation between services, and in understanding how national and local performance can be improved (Grant et al., Reference Grant, Hotopf, Breen, Cleare, Grey, Hepgul and Tylee2014; Goddard et al., Reference Goddard, Wingrove and Moran2015).
Ethnicity is a factor of interest in the context of mental health because there are known variations between ethnic groups in terms of prevalence and outcomes; for example, in rates of detention under the Mental Health Act (Gee and Ponce, Reference Gee and Ponce2010; Karlsen et al., Reference Karlsen, Nazroo, McKenzie, Bhui and Weich2005; Lawlor et al., Reference Lawlor, Johnson, Cole and Howard2012; Simpson et al., Reference Simpson, Krishnan, Kunik and Ruiz2007; Singh et al., Reference Singh, Greenwood, White and Churchill2007; Weich et al., Reference Weich, Nazroo, Sproston, McManus, Blanchard, Erens and Tyrer2004). In a multicultural society, ensuring equity of access, treatment and outcome is an important policy objective. Recent advancements in the understanding of culture on mental health has led to suggestions that ‘modest’ modifications to treatments are necessary to make services culturally applicable (Jankowska, Reference Jankowska2019). IAPT has recognised these ethnic disparities and commissioned the IAPT Black, Asian and Minority Ethnic Positive Practice Guide to address some of the structural issues within IAPT services (Beck et al., Reference Beck, Naz, Brooks and Jankowska2019). However, the dearth of evidence exploring the reasons for such variations in outcomes means it is uncertain if such changes will be enough to solve disparities.
There are several possible sources of variation in outcomes between ethnic groups. A particular focus of this paper is whether demographic differences between ethnic groups may account for some of the differences seen. For instance, social deprivation is known to affect the outcomes of psychological therapies (Clark et al., Reference Clark, Canvin, Green, Layard, Pilling and Janecka2018). Outcomes might vary if people from ethnic minority groups have different baseline levels on average. For example, high levels of deprivation are associated with an increased incidence of CMDs and increased referral rates to IAPT services (Centre for Social Justice, 2011; de Lusignan et al., Reference de Lusignan, Navarro, Chan, Parry, Dent-Brown and Kendrick2011), and socio-demographic factors predict symptom development irrespective of baseline symptom level (Joutsenniemi et al., Reference Joutsenniemi, Laaksonen, Knekt, Haaramo and Lindfors2012).
Analyses by Singh and colleagues found that ethnicity had no independent effect on the odds of being detained under the Mental Health Act (Singh et al., Reference Singh, Burns, Tyrer, Islam, Parsons and Crawford2014) once other predictor variables are controlled for.
In this paper we examine the impact of differences in socio-demographic factors and baseline symptomatology on differences in outcome between ethnic groups.
Method
Participants
From the initial database, we extracted data from patients for all the chosen variables. Patients were included in the analysis if they had available data for all the chosen variables and attended two or more IAPT sessions at which the necessary clinical measures had been completed. Only the first episode of treatment was included for each individual to maintain statistical independence. The key IAPT outcome metrics, recovery and reliable improvement, were calculated prior to data extraction. Descriptors for the 51,762 patients can be seen in Table 1.
Mean of socio-demographic variables, recovery and reliable improvement (RI) by ethnic group
‘Not known’ and ‘Not stated’ refer to groups who did not wish to disclose their ethnic group, or who were not able to identify with the categories provided.
Study design
This was a retrospective observational study of anonymised treatment data collected for routine clinical purposes.
Data sources
IAPT services collect demographic information at treatment entry and patient-reported outcome measures at every treatment session. The data used in this analysis were from routine clinical cases attending nine IAPT services entering treatment between 2009 and 2016. Not all services had started in 2009.
IAPT services collect a wide range of demographic information which includes ethnicity, coded by Office of Population Census and Survey (OPCS) category. A wide range of other demographic data including gender, age and sexuality are also collected (Singh et al., Reference Singh, Burns, Tyrer, Islam, Parsons and Crawford2014). Demographic data are collected at assessment. In the case of variables which may change, for instance employment status, data are collected at each session, or where status changes. A range of standardised measures are collected at assessment and at each treatment session. The key measures are the Patient Health Questionnaire-9 (PHQ-9) and Generalised Anxiety Disorder-7 (GAD-7) scale for measuring depression and generalised anxiety, respectively. Where appropriate, other anxiety measures are used, for example the HAI for health anxiety. These measures are used to calculate outcomes. The Work and Social Adjustment Scale (WSAS), a measure of functioning, is also collected. Outcomes are based on the last available treatment session regardless of whether a planned course of treatment has been completed. People completing two or more treatment sessions are, therefore, included even if they have dropped out of treatment.
Data were extracted directly from the server in anonymised form. No personally identifiable data were extracted, and all data on the system were collected for routine clinical purposes. Date of birth was converted to age in years at the point of extraction so that date of birth was not included in the extracted dataset. Postcode was electronically converted to the Index of Multiple Deprivation (IMD) 2015 score prior to extraction and only the first three characters of the postcode (a very large geographical area) was extracted.
Outcome measures
IAPT uses a cut-off of above 9 on the PHQ-9 and a cut-off of above 7 on the GAD-7 to classify cases as depressed or anxious, respectively (Kroenke et al., Reference Kroenke, Spitzer and Williams2001; Spitzer et al., Reference Spitzer, Kroenke, Williams and Löwe2006). In the case of an anxiety disorder specific measure (ADSM) being substituted for the GAD-7, relevant cut-offs given by the scale developers is used. The WSAS does not have a cut-off (Mundt et al., Reference Mundt, Marks, Shear and Greist2002). Ethnic group is one of the self-reported patient variables recorded as part of IAPT’s routine data.
IAPT uses two binary outcome metrics to assess clinical outcomes: ‘recovery’ and ‘reliable improvement’. Recovery is a metric based on test cut-points and occurs when a patient enters treatment above threshold on either the PHQ-9 or the GAD-7 (or relevant ADSM) and is below threshold on both at the last treatment session for which data are available.
Reliable improvement occurs when a patient’s score on a standardised anxiety or depression measure reduces by more than a specified amount and does not deteriorate by more than a set amount on the other measure at the last session for which the data are available (Clark, Reference Clark2011; Kroenke et al., Reference Kroenke, Spitzer and Williams2011; Spitzer et al., Reference Spitzer, Kroenke, Williams and Löwe2006).
Explanatory variables
All ethnic categories were kept for analysis. The IMD can be calculated from a person’s postcode – the more deprived the individual’s area, the higher the IMD. The WSAS scale measures social functioning and a higher score indicates worse social function. GAD-7 (anxiety) and PHQ-9 (depression) are the scales used to establish severity of CMD.
We included IAPT social determinant variables in this analysis, including age in years at referral, initial severity on the anxiety (GAD-7) and depression (PHQ-9) scales, initial social functioning score (WSAS), gender, long-term physical condition (LTC, calculated as a binary variable of either reporting a long-term physical condition, or not), occupation at assessment (as a binary variable according to whether the patient has a daily meaningful occupation), and ethnicity (as a categorical variable recorded upon registration into the service).
Statistical analyses
In this analysis we examined gender, ethnic group, existing IAPT categories for age in years at referral, and baseline scores on the GAD-7, PHQ-9 and WSAS at assessment. We also created binary variables for whether patients reported to have a LTC (binary value = 1) and whether they had an occupation (binary value = 1), counted as paid employment, full- or part-time students, homemakers, retired individuals, and unpaid volunteers. Those with no occupation included those unemployed and seeking work, long-term sick/disabled receiving benefits, unemployed not seeking work, and individuals not receiving benefits.
First, we ran descriptive data analyses with averages and proportions by ethnic group to identify differences in all the variables chosen. We then tested the chosen variables for intercorrelation and ran a univariate logistic regression to identify those variables which had an effect on recovery and reliable improvement. A prediction plot was calculated to visualise the differences in predicted proportions recovering with their confidence intervals (CIs) for each ethnic group. Finally, we ran a mixed effects logistic regression model, with White British as the reference ethnic category, recovery as the outcome, and subsequently of reliable improvement as the outcome, controlling for all the variables listed and grouping by service to account for clustering. We removed age as an independent variable, as it was not found to be significant in the univariate analyses.
We used the statistical software package STATA 12 and quote 95% CIs, odds ratios (OR) and p-values in the results.
Results
In the nine services, 73,972 patients (81.2% of the sample) had their ethnic group recorded by the service and were included for analysis. However, when cases with missing data on any key variable were excluded, the sample size was reduced to 51,762. The majority of the sample were female (67%) and White British (43.2%) (see Table 1). Over half (58.5%) had an occupation at the beginning of treatment and 18.8% had a recorded LTC.
Ethnic groups varied widely in their socio-economic and social function status (Table 1). For example, patients of ‘Caribbean’ or ‘Any other Black’ background tended to have noticeably higher incidence of LTC and worse WSAS scores. The sizes of the ethnic groups varied from 333 for ‘White and Black African’ to 31,936 for ‘White British’.
Interestingly, the ethnic groups seemed to achieve a lower recovery rate on average. The Pakistani and Bangladeshi populations show considerably higher reliable improvement mean percentages, indicating that although they are not crossing the ‘recovery’ threshold, their morbidity metrics were still improving with treatment. Moreover, discrepancies in rates of recovery for ethnic minority groups relative to White British tended to be somewhat higher than discrepancies for reliable improvement.
A logistic regression using ‘White British’ as the reference category with recovery as the outcome shows that the odds ratio for recovery varies between ethnic groups (see Table 2).
Logistic regression of ethnic group against recovery with White British as the reference group (p < 0.001), n = 51,762
IMD, initial severity, social functioning score, occupation and having a LTC all have a statistically significant impact on recovery (p<0.001). When these variables were entered into the model, the odds ratios associated with identifying as a member of an ethnic group reduced in virtually all black, minority ethnic and refugee (BMER) groups and, for some groups, ethnicity was no longer a significant predictor of recovery (Table 3). The odds ratios for a logistic regression are calculated per unit increase in the predictor variable and can be difficult to interpret. For ease of interpretation, we also calculated them per standard deviation increase for IMD 2015 (0.910), baseline PHQ-9 (0.626), baseline GAD-7 (0.733) and baseline WSAS (0.827).
Mixed effects logistic regression results, group variable: n = 51,762
We then carried out a similar analysis on reliable improvement (Table 4).
Logistic regression of reliable improvement and ethnic group (with White British as the reference group), showing strong statistical significance for a majority of the groups (p < 0.001), n = 73,972
When the same analyses were run using reliable improvement as the outcome variable, the results were similar to those for recovery (Table 5). The odds ratios for membership of ethnic minority groups relative to White British were reduced in most groups relative to the unadjusted model. Again, membership of some ethnic groups was no longer a statistically significant predictor of outcome.
Mixed effects logistic regression results for reliable improvement with White British as the reference ethnic category, group variable: Service, 60,959 observations, 9 groups
Discussion
There are clear differences in recovery rates between White British and most other ethnic groups (Table 1). Those differences are less marked for reliable improvement. The differences between ethnic minority groups and the White British group are reduced for both recovery and reliable improvement once other socio-demographic variables are considered.
The apparent disparity in recovery between populations of different ethnicities is often attributed to cultural differences, which has led to treatment that is culturally adapted for people within BMER groups (Jankowska, Reference Jankowska2019). Although such changes are necessary, these finding suggest that at least some part of the disparity in recovery rates are attributable to general socio-demographic factors and entry level morbidity. These factors affect outcomes in all populations but are present at greater intensity in some ethnic minority populations. For instance, social deprivation levels are higher in BMER groups and greater social deprivation is associated with a reduced probability of reaching recovery. Once some of these socio-demographic variables are considered, ethnicity becomes a weaker predictor of recovery. This is not to deny the importance of culture, simply to point out that it is not the only way in which ethnicity and outcomes can be linked and institutional racism and the disadvantage faced by some of these BMER groups is likely to play a bigger role than cultural differences alone.
Recovery is particularly impacted by level of morbidity at entry because it is a metric which depends on the individual crossing a fixed threshold score. The further they are from that threshold, the more anxious or depressed they are, the larger the change they must make to reach recovery. It is worth recalling that the odds ratio on a continuous predictor variable is the change in probability for each unit increase in the predictor variable and with each unit increase building on the previous level, the effect is compound, not simply additive. The results for reliable improvement are similar, although the impact of starting scores is somewhat smaller and there is a difference in direction of effect for the GAD-7 starting score.
Nonetheless, entering our socio-demographic variables does not eliminate all differences between different ethnic groups in outcome. However, we only entered a limited number of socio-demographic variables. There are others which might have been entered, for instance fluency in English. While this is sometimes thought of primarily as an issue about first language therapy, not speaking fluent English has multiple implications in terms of social life, ability to get well-paid employment, and ease of negotiating everyday life. Some ethnic minority groups have many forced migrants, and we were unable to measure the impact of this as IAPT does not record this variable routinely. The impact of the limited number of socio-demographic variables we entered into our models should not be interpreted as the sum total of all socio-demographic variables which might impact therapy outcomes.
There are some limitations to our available data. For instance, we only had self-reported LTC status. There are limitations in the way that ethnicity is categorised. Grouping all service users from India for example, which has a population of 1.252 billion into one ‘ethnic group’ does not do justice to the complexity of lived experiences and cultural differences. People of Indian descent are the second largest migrant population in the UK (Migration Observatory, 2020), and come from a subcontinent with a rich and complex cultural, economic and social diversity and classification by ethnic group ignores important issues such as whether they are migrants themselves, or the descendants of migrants.
Our findings do not deny the importance of cultural issues and the well-established impact that institutional racism has across public health and medical interventions, they simply show that ethnicity is not just a marker for potential cultural differences and national identity, it is also a marker for multiple disadvantages. Whilst cultural adaptation of CBT services is necessary, it is not sufficient to ensure outcome equity. The broader developments of service change were highlighted by Beck and Naz (Reference Beck and Naz2019) where they outlined the way services must work with BME populations to understand services use, and how to integrate their needs appropriately into service development (Beck and Naz, Reference Beck and Naz2019). Whilst the IAPT report to address structural issues within service provision is promising (Beck et al., Reference Beck, Naz, Brooks and Jankowska2019), it is necessary to take a broad view of the relationship between ethnicity and outcome. To ensure genuine equity, a multi-level approach is needed which addresses both structural racism and socio-demographic inequities within our healthcare and society. This will require greater investment in services in areas with higher percentages of BMER individuals.
Data availability statement
The data used in this analysis were obtained from routine clinical cases attending nine Improving Access to Psychological Therapies (IAPT) services entering treatment between 2009 and 2016. IAPT services collect demographic information at treatment entry and patient reported outcome measures at every treatment session
Acknowledgements
This article presents independent research commissioned by the National Institute for Health Research (NIHR) under the Collaborations for Leadership in Applied Health Research and Care (CLAHRC) programme for North West London. The views expressed in this publication are those of the author(s) and not necessarily those of the NHS, the NIHR or the Department of Health. The authors would like to acknowledge the advice of Professor David Clark, University of Oxford, on the handling and analysis of large IAPT data sets.
Author contributions
Federica Amati: Conceptualization (lead), Formal analysis (lead), Methodology (lead), Writing – original draft (lead), Writing – review & editing (equal); John Green: Conceptualization (equal), Data curation (lead), Formal analysis (equal), Funding acquisition (lead), Methodology (equal), Supervision (equal), Validation (equal), Writing – original draft (equal); Hilary Watt: Data curation (equal), Formal analysis (equal), Writing – review & editing (equal); Sophie Jones: Data curation (equal); Noor Al-Rubaye: Writing – review & editing (equal); Lucy McCann: Project administration (supporting), Writing – review & editing (lead); Geva Greenfield: Writing – review & editing (equal).
Financial support
This article presents independent research commissioned by the National Institute for Health Research (NIHR) under the Applied Health Research (ARC) programme for North West London.
Conflicts of interest
There are no conflicts of interest to declare by the authors. All co-authors have read and approved the manuscript, and there are no financial conflicts of interest to disclose. We certify that the submission is original and has not been submitted to another publication for consideration.
Ethical standards
Authors have abided by the Ethical Principles of Psychologists and Code of Conduct as set out by the British Association for Behavioural and Cognitive Psychotherapies and British Psychological Society. Any previously collected and anonymised data does not necessitate additional research ethics committee review. Research and Development (R&D) approval was given by Chelsea Westminster Hospital R&D Department (reference number for the original research piece: C&W15/076).
Comments
No Comments have been published for this article.