Risk factors for initial appointment non-attendance at Improving Access to Psychological Therapy (IAPT) services: A retrospective analysis

Abstract Background Approximately 1.5 million referrals are made to Improving Access to Psychological Therapy (IAPT) services annually. However, treatment is received in less than half of cases due to ineligibility or non-attendance. The aim was to explore risk factors for non-attendance at the initial two IAPT appointments following referral. Methods An exploratory, retrospective analysis of referral and attendance data from five IAPT services in the North of England. Participants were 97,020 referrals received 2010–2014. Main outcome was attendance at the first two offered appointments (assessment and initial treatment). Results Based on data from two services, 66% of referrals resulted in assessment attendance. Across all five services 57% of patients who attended for assessment subsequently attended the first treatment appointment. The odds of attending an assessment appointment were more than 3 times higher for self-referrals than for GP referrals (OR 3.46, 95% CI 3.27–3.66, p < 0.001). Factors important to treatment appointment attendance following assessment were the service, referral source, presenting problem, and anxiety severity. Conclusion Initial appointment non-attendance is a consistent problem for IAPT services. Specific factors that may support IAPT services to improve non-attendance rates are identified. IAPT indicators of success should take account of non-attendance at initial appointments.


Introduction
In England, primary care mental health services are delivered by Improving Access to Psychological Therapy (IAPT) services which offer talking therapies for common mental health problems. Patients can self-refer or be referred by their General Practitioner (GP) or another healthcare professional, and IAPT services are provided to patients free of charge, through the National Health Service (NHS). It comprises an initial assessment appointment (usually undertaken by phone) to determine whether IAPT treatment is suitable, followed by a course of treatment appointments, usually around seven sessions according to national data (NHS Digital, 2021). IAPT services report treatment outcomes on a session-by-session basis (Clark et al., 2018) and have been shown to provide effective, evidence-based support for people with common mental health problems (Wakefield et al., 2021). Although the treatments offered by IAPT services such as guided self-help and cognitive behavioural therapy, are often beneficial (Barlow et al., 2000;Cuijpers et al., 2010Cuijpers et al., , 2020Norton & Price, 2007;Vernmark et al., 2010), and the service delivery of these is considered effective (Wakefield et al., 2021), non-attendance at early IAPT appointments is a significant problem.
In 2014, 62% of patients referred to IAPT services across England discontinued involvement prior to their first treatment appointment (Delgadillo et al., 2016). The latest annual report for IAPT services show that during the 2020/2021 12-month period, 1.46 million referrals were received by IAPT services in England, from which only 634,649 (43%) individuals attended both an assessment and an initial treatment appointment (NHS Digital, 2021): an attrition rate of 57%. Similar figures have been reported in recent IAPT annual reports (NHS Digital, 2018, 2020. A portion of these people will have been assessed and signposted elsewhere for more appropriate support (National Collaborating Centre for Mental Health, 2018); however, this level of attrition raises concerns about appropriateness of referrals, wasted service resource and unmet patient need. Key performance indicators for IAPT services focus on service delivery and treatment outcomes; however, non-attendance rates for initial appointments are not addressed.
There is a need to understand the factors associated with patient attrition from IAPT. Factors contributing to attrition may vary and there is evidence that different factors may be influential during the early and later stages of a course of treatment (Barrett et al., 2008). Previous research has already highlighted the length of time between referral, assessment appointment and first treatment appointment as important to IAPT appointment attendance (Clark et al., 2018;Davis et al., 2020) and appointment attendance at similar services offering psychological support to people with common mental health problems (Sweetman et al., 2021). To complement this previous work, we considered other factors which have been reported to be influential in mental health appointment attendance, within the context of IAPT referral data. We focussed on non-attendance at the initial two contacts with IAPT services: an assessment appointment, and the first treatment session offered.

Aims
(1) To ascertain which characteristics increase the risk of non-attendance at an assessment appointment in five IAPT services in the north of England.
(2) To ascertain which characteristics increase the risk of non-attendance at a first treatment appointment in five IAPT services in the north of England, among patients who had attended the assessment appointment.

Design and Data Sources
This was a retrospective analysis of routinely collected clinical data from psychological services. Data from five IAPT services located in the north of England were obtained through the Northern IAPT Practice Research Network, a multi-service collaboration between IAPT services and academic partners (Lucock et al., 2017). The dataset included individual patient data representing 97,020 referrals made between January 2010 and December 2014.

Measures
Variables included those routinely collected by the services to comply with national data reporting standards: method of referral, IAPT service, gender, age, and ethnicity. Each patient's home postcode was linked to the English Index of Multiple Deprivation (IMD) (Smith et al., 2015). The IMD is a measure of relative deprivation for small geographical areas in England (neighbourhoods with an average of 1,500 residents in each), referred to as Lower-layer Super Output Areas (LSOA). IMD data were available in two forms: a scale variable where higher numbers represent higher levels of deprivation, and as deciles which rank geographical areas across England (rank 1 = 10% most deprived areas, rank 10 = 10% least deprived areas). Attendance at each offered appointment was recorded and additional information was collected at each attended appointment. Following assessment, information about the presenting problem, severity of depression symptoms via PHQ-9 (Kroenke et al., 2001), anxiety symptoms via GAD-7 (Spitzer et al., 2006), employment status and decisions about subsequent appointments were recorded.

Ethical Approvals
Data collection for this project was reviewed and approved by North East -Newcastle & North Tyneside 2 Research Ethics Committee (REC reference: 15/NE/0062) and the University of York Department of Health Sciences Research Governance Committee.

Data Preparation
All analysis was conducted in Stata v16 (StataCorp, 2019).

Analysis Plan
Patient identifiers were not included in the dataset, and it is possible that the data contain multiple referrals relating to the same patient. However, for the purposes of this analysis, referrals were considered to be independent. Separate analyses were conducted to examine attendance at an initial assessment appointment and treatment initiation after an assessment, since additional variables were available for individuals who attended the initial assessment. Data relating to all referred individuals (including those who did not attend for assessment) were only available for two of the IAPT services; these were included in the analysis of risk factors associated with non-attendance of the assessment appointment. All five IAPT services provided data on patients who attended the initial assessment, regardless of whether or not they continued to treatment. These were included in the analysis investigating risk factors associated with non-attendance at the initial treatment session. Logistic regression analyses were applied firstly using all available variables as covariates and then using backwards selection of predictors (Nathanson & Higgins, 2008;Peng et al., 2002). The latter models aimed to retain statistically significant patient-level variables as fixed effects to ascertain which factors were associated with increased odds of non-attendance. Variables considered in the analysis of attendance at assessment following referral were referral source, IAPT service, gender, age in years, ethnicity, and IMD score. Additional variables included in the analysis of treatment initiation (initial treatment appointment attendance) were employment status, primary presenting problem, and symptom severity measures (PHQ-9 and GAD-7). A significance level of p < 0.1 was used to determine the categories and variables to be removed from the models during backwards selection, which proceeded until the final fully-adjusted model only included predictors with a significance level of p < 0.05. Service was included as a fixed, rather than a random, effect due to the small number of services included; therefore, interpretations of findings will necessarily be limited to the services within this dataset (Allison, 2009). The overall effect of each categorical variable was tested using the Wald Chi-squared test; those with a p value greater than 0.05 were not considered to be a significant predictor of the outcome. The Hosmer-Lemeshow Chisquared test was used to assess goodness of fit of the logistic regression models.

Missing Data
Patterns of missing data were explored by variable. Pearson's Chi-squared test of independence was used to ascertain whether missing data points within each variable were different to values present for the same variable when compared to the outcomes of attendance at assessment and initial treatment appointments. The number of missing data points per referral observation was also evaluated to assess whether the number of missing values within referrals was associated with different outcomes for attendance at assessment or initial treatment appointments. Thirdly, missing data were considered by service to assess differences in reporting patterns across services. Finally, after "missingness" within the original dataset had been explored, missing values were replaced through multiple imputation (MI) (Rubin, 1976) by chained equations (White et al., 2011). MI (burnin iterations = 10, number of imputations = 10) was used to replace missing values within the explanatory variables of ethnicity, employment status, a condensed version of the presenting problem variable (phobic anxiety disorder, other anxiety disorder, obsessive compulsive disorder, stress or adjustment disorder and somatoform disorder were collapsed into one category labelled other CMH disorders), PHQ-9 and GAD-7 using the logit function, and IMD using the regress function. The presenting problem variable was collapsed to reduce the number of categories within the imputation modelling as large numbers of categories caused problems with convergence. Age and service were included in the imputation command to provide additional information which would support predictions of missing values. Ten iterations were selected to allow successful convergence of the model; variable characteristics were proportionally Psychotherapy Research 537 the same across different imputed datasets and reflect the original data. Analyses were rerun on these multiply imputed datasets and the parameter estimates were pooled using Rubin's rules.

Description of the Dataset
A total of 45,332 referrals were included in the examination of risk factors associated with assessment appointment non-attendance (Services A and C only), and 97,020 were included in the assessment of risk factors for non-attendance at initial treatment appointment (Services A-E) (see Table I).
Characteristics of people referred are presented in Table II, stratified as: did not attend any appointments, attended only an assessment appointment, or attended an offered treatment appointment following an assessment. The number of people considered ineligible for treatment was low (<5% of those referred) for most services with the exception of Service D where 32% of referrals were considered ineligible. In Service D, 83% of referrals were made by a GP or other professional (see Table I), indicating the potential for reducing inappropriate referrals through closer liaison with referring professionals. Among those considered eligible at assessment, treatment appointment attendance ranged between 59% and 79% across services. Overall, 50% of the referrals in this dataset resulted in an assessment and attendance at a first treatment appointment.

Non-Attendance at Assessment Appointment
In total, 29,779 (66%) referrals to services A and C were recorded as attending for an assessment appointment. Characteristics of people referred to services A and C, presented by whether or not they attended for their initial assessment are shown in Table III. A logistic regression model was run to assess the association between all variables entered and attendance at assessment appointment, see Table IV. Based on a p-value of <0.05, this model indicated that referral source may be important to consider when assessing risk of non-attendance to assessment appointments. The odds of attending an assessment appointment were more than three times higher for people who self-referred than for those who had been referred by a GP (odds ratio (OR) 3.44, 95% CI 3.24-3.65, p < 0.001). No significant differences were noted in assessment attendance between people who were referred by a GP and those who were referred by another professional.
Age, ethnicity, and IMD score were also significantly associated with assessment attendance; however, the odds ratios were small indicating modest effect sizes. People who attended tended to be older than those who did not attend assessment appointments (OR 1.017, 95% CI 1.015-1.019, p < 0.001). When compared with White people, individuals from Mixed, Black and Other ethnic groups were significantly less likely to attend assessment appointments (OR for Mixed 0.70, 95% CI 0.60-0.83, p < 0.001, Asian OR 0.87, 95% CI 0.75-1.00, p = 0.06, Black OR 0.80, 95% CI 0.66-0.95, p = 0.01, Chinese OR 0.80, 95% CI 0.46-1.39, p = 0.42, Other OR 0.63, 95% CI 0.50-0.80, p < 0.001). Higher IMD scores (indicating more deprived areas) were significantly associated with assessment appointment non-attendance (OR 0.990, 95% CI 0.989-0.992, p < 0.001). No significant differences were observed for gender. Service was not included in this model. Univariate odds ratios for each predictor on assessment attendance are also included in the supplementary file.
Following the initial regression analysis, a stepwise regression using backwards selection was undertaken. As in the full regression model, there were 30,501 observations included in the backwards selection model of services A and C. Gender (p = 0.69), referral by GP or other professional (p = 0.47) and ethnic group of Chinese relative to White (p = 0.41) were removed from the model as no significant differences were found. As in the full model, the stepwise regression analysis indicated that method of referral, age, ethnicity and IMD score were important to assessment attendance. Full details can be found in the supplementary file.

Psychotherapy Research 539
Non-Attendance at Initial Treatment Appointments A logistic regression model was run to assess the association between all variables entered and whether individuals attended an initial treatment appointment, see Table V. This indicated that the service people were referred to, method of referral, presenting problem and severity of anxiety may be important to consider when assessing risk of nonattendance to initial treatment appointments. Ethnicity and severity of depression were not found to be associated with initial treatment attendance. In this model people referred to service A were significantly more likely to attend a treatment appointment following an assessment than people referred to services B or D (OR for service B vs. A 0.79, 95% CI 0.74-0.84, OR for service D vs. A 0.40, 95% CI 0.37-0.42). Individuals referred to services C and E were significantly more likely to attend treatment than those referred to service A (OR for service C vs. A 1.39, 95% CI 1.25-1.54, OR for service E vs. A 2.52, 95% CI 2.37-2.69). This indicates that the odds of attending a treatment appointment were substantially lower (60%) for people referred to service D, and 2.5 times higher for people who were referred to service E, compared to those referred to service A.
Individuals who were referred by another professional were significantly more likely to attend a treatment appointment than those referred by a GP (OR 1.63, 95% CI 1.52-1.75, p < 0.001). This indicates that the odds of attending a treatment appointment were 63% higher for those who had been referred by another professional compared to people who had been referred by a GP. Within this analysis, no significant difference in treatment attendance was noted between those who selfreferred to IAPT service and those referred by a GP.
As in the full model, the stepwise regression indicated that service, source of referral, presenting problem and severity of anxiety were important to treatment attendance. Gender, age, IMD and some categories of employment status were also found to be statistically significant. Full details can be found in the supplementary file.

Missing Data
The most common single variable in the analysis with a missing datapoint was ethnicity, affecting 18,146 cases (19%). Minimal missingness was noted for IMD, referral source and gender. Chi-square tests indicated that missing datapoints for gender were not related to attendance at assessment; however, missing data for ethnicity, IMD and referral source were related to assessment outcomes. Higher levels of missingness were noted in each of these variables for individuals who did not attend an assessment appointment. This was especially apparent for ethnicity, where 54% of those who did not attend an assessment appointment had missing data for ethnicity compared with 12% missingness for those who did attend an assessment. In relation to the initial treatment appointment, missingness within gender, IMD decile and referral source were not found to be related to attendance. Missingness within ethnicity, employment status, presenting problem, PHQ-9 and GAD-7 were related to initial treatment appointment attendance: 16% of people who did Psychotherapy Research 543 not attend for treatment had a missing value for ethnicity compared with 10% of those who did attend the initial treatment appointment; 12% of those who did not attend were missing data for employment status compared to 8% for those who attended treatment; and 21% of people who did not attend for treatment had a missing value for their presenting problem compared with 11% missingness for those who did attend the initial treatment appointment. Missingness for PHQ-9 and GAD-7 were similar with 10% (PHQ-9) and 11% (GAD-7) of people who did not attend treatment having missing data, compared to 4% missing for those who did attend (both PHQ-9 and GAD-7). Data reported by services was examined; with the exception of gender, significant between service differences in missing data were observed for each variable included in the primary and secondary analysis.
In the MI analysis, there were 45,327 observations included in the logistic regression model for assessment attendance using imputed data (Table VI). The pattern of covariates remained consistent with the complete case analysis of assessment attendance. The source of referrals was a strong predictor of assessment appointment attendance in this model; those who self-referred and those referred by other professionals had significantly higher odds of attending an assessment appointment than those referred by a GP (self-referrals OR 2.73, CI 2.61-2.85; professional referrals OR 1.11, 95% CI 1.04-1.19). All other variables were either associated with increased likelihood of attendance at assessment appointments with odds ratio values close to 1, or were not significant within the model. There were 79,692 observations included in the model for initial treatment appointment attendance using imputed data (Table VII). Again, the pattern of covariates remained consistent with the complete case analysis of initial treatment appointment attendance; the service people were referred to, the source of the referral and the problem people were referred for were considered important to treatment attendance, while ethnicity and depression symptom severity were not considered important to treatment attendance.

Discussion
This study provides an insight into the levels of access and support for common mental health problems in the UK using a large dataset of referrals made to five IAPT services based in the North of England between 2010 and 2014. Overall, twothirds of referrals resulted in assessment attendance (based on two services) and across all five services, just over half of those who were assessed subsequently attended a first treatment appointment. The main risk factor for non-attendance at assessment appointments was found to be referral by GP as opposed to self-referral. Risk factors associated with non-attendance at first treatment appointments were being referred by GP, having a depressive disorder or not meeting the criteria for a common mental health problem, and having lower anxiety symptoms. In addition, significant differences in assessment attendance were related to ethnicity, age and IMD score. Gender, age, IMD and some categories of employment status showed significant differences in relation to first treatment appointment attendance.
Historical data were used for this analysis; however, the level of non-attendances found in Table IV. Full logistic regression of those who attended an assessment appointment following a referral to either IAPT service A or C.  Digital, 2018), In 2018-2019 of the 1.6 million referrals received, 1.09 million (68%) attended at least one appointment and 582,556 (36%) attended at least two appointments (NHS Digital, 2019). In 2019-2020 1.69 referrals were received; 1.17 million (69%) resulted in attendance to at least one appointment, 606,192 (36%) of these referrals resulted in attendance to at least two appointments (NHS Digital, 2020). In 2020-2021 1.46 million referrals were received, 1.02 million (70%) were associated with attendance to at least one appointment and 634,649 (43%) were associated with attendance to at least two appointments (NHS Digital, 2021). This indicates that non-attendance to these initial two appointments is a consistent problem for IAPT services which requires some form of intervention to encourage improvements. IAPT annual reporting includes details of service performance relating to service delivery and treatment effectiveness. At present, none of the key performance indicators relate to initial non-attendance; emphasising the importance of attendance at these appointments may support services to address this issue.

Psychotherapy Research 545
The number of services included in this dataset was small (n = 5) which meant that hierarchical factors that may have influenced non-attendance were not able to be explored within this study (Greenland, 2000). This is particularly relevant to the analyses pertaining to assessment appointment attendance, where only two services were able to be included; this is an unfortunate limitation of the data available for this work. In addition, all of these services were based in the north of England; therefore, it is not known whether these findings are relevant to IAPT services in other areas. Additionally, data within this dataset were not identifiable; consequently, it was not possible to control for individuals associated with multiple referrals within the analyses. Although most of the variables included in the analysis were considered objective, the presenting problem relies on the experience of the assessing IAPT practitioner and service-specific protocols to determine appropriate support for people referred (National Collaborating Centre for Mental Health, 2018). There is some evidence to suggest that the accuracy of problem descriptors reported by IAPT services is improving (Saunders et al., 2020); however, there is a lack of research exploring the overall reliability of this variable.
This study has identified specific characteristics that are considered to affect non-attendance at these two distinct time-points. From this, services can explore appropriate strategies to reduce nonattendance. Services that identify non-attendance at assessment appointments as problematic may look to their source of referrals. With the odds of attending an assessment many times higher for people who had self-referred compared to those referred by a GP this could indicate that where people are able to self-refer, the current referral and appointment scheduling processes are acceptable. IAPT services which only accept professional referrals may wish to consider introducing self-referrals. Those services which already accept self-referrals may decide to pilot methods to promote self-referrals. Where IAPT services accept self-referrals, instances of people being referred by a GP may indicate a different approach from IAPT services is necessary to engage people. Conversations between IAPT services and GPs could help to identify whether additional strategies are required to support this group of people to attend initial appointments (Thomas et al., 2020).
In line with the data from the analysis of assessment appointment attendance, high levels of nonattendance at treatment appointments may indicate that different approaches are needed to support early appointment attendance for people referred by GPs when self-referral is an option. Further research exploring the different needs of in people who self-refer and those referred to IAPT by a GP would be advantageous. Interestingly, people who had been referred by a professional other than a GP were most likely to attend for treatment after attending an assessment appointment. Further exploration of this finding could help to determine whether the nature of the professional referral, or the processes involved in the referral from these professionals, could explain why people referred by professionals other than a GP are more likely to attend an initial treatment appointment.
Detailed comparisons between the referral pathways for services with higher and lower attendance rates for treatment appointments may help to identify service-level practices associated with increased early appointment attendance; for example, the odds of attending a treatment appointment at service E were much higher compared to referrals to service A. Clarifying the scope of the service prior to referral, for all referrers (people requiring support and professionals), could help to prevent inappropriate referrals. In line with this, the odds of attending treatment decreased by 66% for those who did not meet criteria for a common mental health problem, and by 44% for those with other less common problems (eating disorder, alcohol related mental or behavioural disorder, bipolar affective disorder, non-organic sleep disorder, and not specified), compared to individuals with a depressive disorder. Ensuring that IAPT service information incorporates details for other local organisations which offer support for problems outside of the IAPT remit would enable people to contact the appropriate services to meet their needs.
In the wider mental health literature, the finding of depressive disorder being associated with nonattendance to mental health appointments is in line with other research; a survey of primary care patients found that depression was associated with more perceived barriers to accessing mental health support than non-depressed patients (Mohr et al., 2006). In addition to having more perceived barriers to attending treatment, theories relating to the level and type of motivation associated with treatment appointment in people with depression may help to explain nonattendance rates for this group (Binnie & Boden, 2016;Trew, 2011). Also in line with our findings, anxiety has been associated with attendance to mental health appointments where other diagnoses were associated with non-attendance or drop out from both mental health services (Bados et al., 2007) and also with increased healthcare use more generally (Horenstein & Heimberg, 2020). Although these studies do not specifically relate to IAPT service non-attendance, the similarities in findings suggest that the nature of the presenting problem is important to initial appointment attendance rates. Our finding that the people with more severe anxiety symptoms are more likely to attend initial appointments is also in line with previous research findings; where people consider that their mental health problems are not severe or could resolve without additional support, non-attendance is considered more likely to occur (Bados et al., 2007;Bruwer et al., 2011;Greeno et al., 1999;Sweetman et al., 2021). Many of the other characteristics included in these analyses showed statistically significant differences in relation to attendance at assessment or initial treatment appointments. These characteristics reflect an individual's gender, age, ethnicity, whether they live in an area of social deprivation, employment status and the severity of anxiety symptoms. These findings were consistent whether missing data were excluded from the analyses or replaced using an imputed dataset. Increasing the awareness of higher attrition rates amongst certain population groups may be useful for considering adjustments to IAPT services and the patient pathway.
Other factors likely to affect initial appointment attendance include delays between a referral being made and initial appointment dates. The length of time between referral, assessment appointment and first treatment appointment has been highlighted as important specifically for IAPT services (Clark et al., 2018;Davis et al., 2020) and services offering psychological support for common mental health problems more generally (Sweetman et al., 2021); however, this was not explored within this sample. Individual reasons for non-attendance including perceptions of mental health symptoms and psychological therapies, support from network members, previous contact with similar services and logistical issues around appointment attendance were not able to be explored using this dataset but have been shown to influence attendance at initial appointments at similar primary care mental health services (Sweetman et al., 2021). Further work is needed to assess whether and how these individual factors influence non-attendance at initial appointments to IAPT services.
Based on data from two services, we showed that only 39% of referrals resulted in attendance to both assessment and first treatment. Based on data from 5 services, we showed that only 57% of people attended their first treatment appointment after attending an initial assessment. With the proportion of inappropriate referrals found to be minimal for four of the five services in this sample, this level of attrition is a significant problem.
These findings suggest a need for changes to IAPT delivery and the reporting of IAPT effectiveness: indicators of success of the service should take account of self-selected patient attrition. It may be helpful for individual IAPT services to monitor rates of non-attendance and assess the factors associated with it; interventions to increase attendance could be targeted as indicated. There is also a need for further research to (1) understand the individual reasons behind non-attendances which could inform flexibility in service policies, (2) explore the most effective and acceptable self-referral pathways to IAPT care and (3) investigate differences between services in the referral pathway which may influence initial non-attendance rates.

Disclosure Statement
No potential conflict of interest was reported by the author(s).

Supplemental data
Supplemental data for this article can be accessed online at https://doi.org/10.1080/10503307.2022.214 0616.