The costs of bipolar disorder in the United Kingdom

Abstract Objectives To estimate the individual cost and population‐level economic burden of Bipolar Disorder (BD), and explore the impact of clinical and sociodemographic factors on costs in the United Kingdom. Methods Annual UK health care, social care and societal cost data were collected from a prospective cohort of 91 BD patients using digital monitoring of symptoms. Costs (in £) were calculated for the year of resource use collection (2010–2011) and main results inflated to year 2018–2019. A Generalized Estimating Equation framework was used to investigate individual factors influencing costs. An economic burden estimate was derived by multiplying the mean annual cost per patient with literature‐based population prevalence. Results The average annual cost of BD per patient was £12,617 (SE = ±£1085) or £14,938 (SE = ±£1281) at 2018–2019 prices with 68% of the total costs attributed to lost productivity and informal care, 31% to health care costs, 1% to private out‐of‐pocket expenses, and 0.5% to social care costs. A unit increase in average levels of depressive or manic symptoms were associated with 7% and 11% higher societal costs, respectively. The estimated annual prevalence of BD in the United Kingdom was 0.8% resulting in a population‐level economic burden estimate of £5.1 billion for 2010–2011 or £6.43 billion for 2018–2019. Conclusions BD is a disease of substantial costs in the United Kingdom with the majority of the economic burden falling outside the health care system in the form of productivity losses and informal care. These costs highly correlate with patient outcomes highlighting further needs for improved treatment efforts into functionality.

mechanism, although there are several reasonably effective treatments for the acute illness and for preventing relapse (Geddes & Miklowitz, 2013;Kroon et al., 2013;Perlis et al., 2006). BD leads to significant impairment of health-related quality of life (HRQOL) and wellbeing along with detrimental emotional, financial and health implications not only for the patients, but also their families and the wider society (Ishak et al., 2012). However, there is less knowledge of the total disease burden associated with BD in the United Kingdom. Estimating the costs of BD can provide valuable insights into its relative burden and cost distribution along with the opportunity to explore factors that have the greatest impact on health, social care and societal resource utilization. This evidence may be critical in raising awareness of the disease or aspects of the disease, thereby contributing to discussions and decisions around efficient resource allocation (Larg & Moss, 2011).
Few studies have comprehensively assessed the economic burden of BD in Europe, and only three studies have attempted to estimate it in the United Kingdom according to the latest international review (see Appendix Table A1) (Jin & McCrone, 2015).
The two previous economic burden assessments for BD in the United Kingdom (Das Gupta & Guest, 2002;McCrone et al., 2008) are based on data that are more than 20 years old, whereas the more recent study by Young et al. (2011) for the year 2007-2008 only considered health care-related costs, which previous studies showed to represent only a minor fraction of the total economic burden. All three UK-based studies gave very different health care cost estimates, most likely due to the use of divergent prevalence estimates for BD and the variance in the applied study methodologies (see Table 1).
Over the past decade there has been an emergence of new therapies (e.g., second generation antipsychotics), modalities of care (e.g., adjunct psychotherapies) and service delivery options (e.g., collaborative care, reduced admission beds) (Geddes & Miklowitz, 2013). Notwithstanding the high variation of the results of the three existing UK-based cost-of-illness studies measuring the population-level economic burden caused by BD owing to their methodical heterogeneity, it is likely that all of them actually underestimate the current economic burden of BD in the United Kingdom on the counts of being based on older and outdated therapy regimens and care pathways. Therefore, a more up-to-date cost-of-illness study both in terms of data and methodology following current good practice recommendations is required (Kleine-Budde et al., 2014).
This study seeks to (a) assess the cost of BD over a 1-year period through prospectively following-up a well-defined cohort of community BD patients in the United Kingdom, (b) explore the main clinical and sociodemographic factors that drive variations in the costs associated with BD, and (c) estimate the population-level economic burden for the United Kingdom.

Study population
The study used data obtained from the OXTEXT-2 substudy which was part of the OXTEXT research programme (RP-PG-0108-10087) and followed a cohort of 91 BD patients over a 12-month period in 2010-2011. Patients were invited to take part between May 2010 and June 2011 from the larger OXTEXT-1 cohort study which was a pragmatic longitudinal cohort of patients with a confirmed diagnosis of BD who attended psychiatric clinics in Oxford Health NHS Foundation Trust (OHFT) and were monitored with a user friendly SMS or web-based system called "True Colours" (TC) (Goodday et al., 2020).
Eligible criteria were: males and females aged over 16 years old, with a primary diagnosis of BD according to the DSM-IV-R criteria, who were willing and able to provide informed consent to participation in the study. The study received ethical approval by the local Research Ethics Committee (Oxfordshire REC A, Reference: 10/H0604/13).
Clinical status of patients was measured using the five-item Altman Self-Rating Mania (ASRM) scale and 16-item Quick Inventory of Depressive Symptoms-Self Report (QIDS-SR) scale. The ASRM scale (scores range from 0 to 20) consists of five questions marked between 0 and 4 with increasing severity. In the literature, cut-off point of less than or equal to 5 is indicative of "being not in a manic state," whereas a score of greater than 5 indicates a "state of mania" (Altman et al., 1997). The QIDS-SR consists of 16 items each scored between 0 and 3. Scores range from 0 to 27 and can be categorized as (a) no depression: 0-5, (b) mild depression: 6-10, (c) moderate depression: 11-15, (d) severe depression: 16-20, (e) very severe depression: 21-27 (Rush et al., 2003).

Data collection
The data on the use of health and nonhealth care resources were collected using self-reported patient questionnaires based on 3-month recall at four different time points: 0-to 3-, 4-to 6-, 7-to 9-, and 10-to 12-month follow-ups. In addition to baseline information on sociodemographics, disease status and illness duration, data were collected on contacts made with primary, secondary, tertiary and social care; psychiatric medication; patients' out-of-pocket expenses (e.g., private dietician, acupuncture); and productivity losses. Data on distance travelled to psychiatric health care providers and the time informal carers (e.g., family members and friends) spent providing unpaid care and support was also collected for each study participant.

Cost estimation
We used both health and social care, and societal perspectives in our  Appendix Table A2).
Lost productivity costs were estimated using the human capital approach (HCA). Days off work were multiplied by the average daily cost of sickness absence in the United Kingdom (CMH, 2007;NICE, 2009;Tarricone, 2006). These costs were estimated only for those who were employed or self-employed (n = 53, 58.2%). Informal care costs were estimated by multiplying the average hourly salary derived from the average annual gross UK wage rate with the number of hours of unpaid work (Bovill, 2013;Faria et al., 2012).

Statistical analysis
A p-value of ≤ .05 was used in all hypothesis tests of statistical significance. All statistical analyses were undertaken using STATA 12.0 (Stat-aCorp, College Station, TX, USA).
Multiple imputation (MI) by chained equations was used with age, gender, type of BD, duration-of-illness, concurrent clinical measures, and the value of the missing variable at baseline as covariates (Schafer & Graham, 2002). The MI by chained equations procedure ("MI impute") with predictive mean matching to account for skewed distribution in cost categories was applied to perform 100 imputations (MacNeil Vroomen et al., 2016).
To determine the influence of major sociodemographic and clinical characteristics such as age, gender, type of BD diagnosis, duration of illness and clinical status on overall costs and other major cost categories, Generalized Estimating Equation (GEE) was implemented on the fully imputed dataset. A multivariate regression model with gamma distribution and log-link function was used to address skewed cost data and an exchangeable covariance matrix was found appropriate to account for the correlation between repeated covariates and their influence on the primary outcomes which were measured on multiple occasions (Raikou & McGuire, 2012

Population-level economic burden
The cost-of-illness of BD in the United Kingdom was estimated as the product of the total mean annual societal cost per BD patient multiplied by the population prevalence of BD. In the absence of a UK

Sensitivity analyses
A series of one-way sensitivity analyses were performed to explore how uncertainties in key input parameters impact the economic bur-
Baseline characteristics of the full OXTEXT-2 study sample (n = 91) alongside the characteristics of the complete cases (n = 48) are provided in Table 1. The baseline characteristics of the included participants were similar to those of the original OXTEXT-1 cohort ( Table 1).
Most of the participants were female (58/91, 64%), the baseline aver- The mean baseline QIDS score for depression was 8.2 (SE = ±0.51), the mean baseline ASRM score was 3 (SE = ±0.34). No episodes of selfharm, completed suicides or deaths were recorded during the study period.

Cost estimates
The mean annual societal cost of BD per patient was estimated at

Health care costs
Primary care costs included visits to/by general practitioners (GP) and general practice nurses. GP costs accounted for 88% of the total primary care costs. The mean primary care cost per patient was £253 Overall, hospitalization costs accounted for 16% of the total annual economic burden per patient and formed the largest health care cost subcategory.

Psychiatric medication costs
Mean annual psychiatric medication costs were estimated at £859 (SE = ±£65, 2018-2019 prices: £1014) per patient, the second largest health care cost subcategory contributing to 7% of the total annual economic burden per patient.

Out-of-pocket patient costs
Patient costs were out-of-pocket expenditures spent on services offered by private health service providers (e.g., dieticians, chiropodists, private physiotherapists, etc.) and travel costs to mental health facilities. The average annual out-of-pocket costs per BD patient were £111 (SE = ±£20, 2018-2019 prices: £131).

Indirect costs (lost productivity and informal care)
Major

Determinants of costs
Results of the GEE analyses are presented in Table 3. Mean annual costs were significantly associated with mean QIDS and ASRM scores over 12 months. A single unit of increase in the QIDS score was associated with 7.7% (p = .012) increase in the mean annual societal cost and for a single unit increase in the ASRM score, the increase in mean annual societal costs was 11.6% (p = .022). When health care costs were considered, only the association with the severity of mania F I G U R E 1 Distribution of the total societal costs and health care costs of BD by major categories remained significant at the 5% level; a unit difference in the mean ASRM score was associated with an increase in the mean annual health care costs of 11.5% (p = .03). Costs due to lost productivity and informal care were more closely associated with the severity of depression; a single unit of QIDS score increase was associated with an increase in annual indirect costs of 9.2% (p = .036).

Population-level economic burden
A total of 913 citations were retrieved following the initial search, out of which, 26 studies were included for abstract review. Eventually, seven studies were deemed as eligible and were included in a randomeffects meta-analysis to derive a pooled estimated for the 12-month period prevalence of BD relevant to the United Kingdom for the study period (see Appendix Figure 1).
The characteristics of included studies are provided in Table 4.

Type of prevalence
Faravelli et al.

Sensitivity analyses
We conducted a series of one-way sensitivity analyses and a PSA to

DISCUSSION
The Overall, our bottom-up economic burden study confirmed that the majority of the societal costs of BD fall on those who experience the illness and their families, rather than the health care system. Health care expenditure comprises slightly less than a third of overall costs. This is consistent with the existing literature that has highlighted the substantial burden of lost productivity in individuals with BD (Jin & McCrone, 2015;Modini et al., 2016). Individuals with BD consider employment or re-employment as one of the major elements in their recovery process (Drake et al., 2012). It is, therefore, not only important to investigate the cost-effectiveness of long-term management strategies, but also interventions that improve the functioning of individuals who are at work.
Economic burden studies are difficult to compare as they vary in methodology, study setting, and type of costs included. In the United Kingdom, Das Gupta and Guest's (2002) retrospective study used a prevalence estimate of 0.5% adopted from the narrative review of the literature by Bebbington and Ramana (1995), but it was significantly different from the originally reported population-based estimate of 1%-1.5%. Direct costs in their study accounted for 14% and indirect costs (summarizing costs due to excess unemployment, absenteeism, and suicide) accounted for 86% of total costs in contrast to 31% for direct costs and 68% for costs due to lost productivity and costs for informal care observed in our study. This is likely to be due to the top-down approach they adopted compared with the bottomup approach used in our study. The increased proportion of indirect costs derived in their study was driven by costs assigned to lost employment (85% of indirect costs) followed by costs due to premature mortality (suicides). The estimates for lost employment were obtained from nonpeer-reviewed reports and assumptions. However, the high proportion of individuals who were unemployed but could work (26%) assumed in Das Gupta and Guest's study (2002) seems to be at odds with much lower rates observed in other relevant studies (Marwaha et al., 2013;McCrone et al., 2008). Moreover, their estimates from 1997 to 1998 included health care resource use of patients who suffered from unipolar psychosis and schizoaffective disorder and this may not be reflective of current practice. Further, we had no observed suicide or other death in our study. Inclusion of mortality-based production losses incurred in future is generally controversial for prevalence based cost-of-illness studies as it does not relate to other timedefined costs that are usually measured over a year (Jin & McCrone, 2015).

McCrone et al. (2008) modeled societal costs of BD in England by
combining health care costs derived from a sample of 103 BD-I patients participating in a randomized controlled trial (RCT) of cognitive behavioral therapy in England, informal care costs adopted from a psychosis study, and productivity losses adjusted from Das Gupta and Guest's study (2002) along with prevalence data from the United States (Lam et al., 2003;McCrone et al., 2008). Even though, the broad distribution of costs with regard to proportion of direct (31%) and indirect (69%) costs in McCrone et al. (2008) is consistent with our present study, the reported share of hospitalization costs in health care-related costs (14%) was much lower compared to the result of our present and other similar studies (Ekman et al., 2013;Tafalla et al., 2010;Young et al., 2011). A possible explanation for this is that in assessing service costs, they combined bottom-up annualized costs based on 3month health care resource use for BD-I patients obtained from a RCT with top-down estimates of hospitalization (Hospital Episode Statistics data) and residential care for BD-related conditions. Moreover, for the majority of resource utilization categories, their estimates were derived from a RCT that was undertaken in 1998-1999 with stringent inclusion and exclusion criteria (excluding patients with recent history of active episodes) (Lam et al., 2003). Such samples may not be representative of individuals with BD in the community.
The most recent research assessing costs of BD in the United Kingdom was a top-down study undertaken by Young et al. (2011), that included only health care costs. They reported that the total hospitalization cost was £207 million corresponding to 60% of all health care costs; this translated to £2370 per BD patient. In comparison, in our present study, hospitalization-related costs were approximately £799 A more recent literature review for studies from the United States by Bessonova et al. (2020)  Overall, the findings of our study are well in line with the large body of international studies on the economic burden of BD when considering the substantial heterogeneity in methods and data types, as well as the numerous differing combinations of cost categories and known system differences.
Further, our present study used True Colours, a method for routine data collection based on patient reported symptoms that is regularly and frequently updated. In an earlier study the system showed high compliance and no impact on mental health service costs except somewhat increased psychiatric medication costs (Simon et al., 2017). Repeated patient reported clinical outcomes may also provide a metric for modeling the economic effect of treatment benefits from RCTs.
A final minor point relates to the validity of the ASRM. Self-rating of mania is often regarded as suspect of bias and potentially confounded by positive mood. In fact, a unit difference in the ASRM was associated with 11% higher illness costs. This implies that the scale does measure a clinically meaningful and impairing effect on health.

Strengths and limitations
This is the first economic burden study that used digitally captured, real-world, self-reported population-based data of BD patients Due to the lack of epidemiological studies of BD in the United Kingdom for the year of resource use collection, prevalence estimates were derived from a random-effects meta-analysis of population based epidemiological studies of BD in Europe. This yielded a mean prevalence of 0.8% (95% CI = 0.43%-1.3%), which is consistent with a more recently published study that reported the prevalence of BD in the United Kingdom using Bayesian meta-regression modeling to be 0.8% (95% CI = 0.7%-0.9%) in females and 0.6% (95% CI = 0.5%-0.7%) in males (Ferrari et al., 2016). An estimate of the UK prevalence was made in Adult Psychiatric Morbidity survey of 2014 (Marwaha et al., 2016). Using a relatively crude self-rated screening instrument, the overall prevalence was 2%, and 0.6% had had the diagnosis confirmed by a health care professional, the cohort most similar to our study. The latter figure is comparable with the European average and illustrates the fact that most patients who screen positive for bipolar disorder are not in active clinical care. By limiting the study to patients with diagnosed and treated BD-as we do in the present study-one is looking only at the tip of the iceberg when considering the burden and cost of bipolar disorder. Taking into consideration the undiagnosed and hence unobserved cases, the complete cost-of-illness of BD is likely substantially higher.
We want to caution that our study results need to be interpreted in light of some important limitations. First, the relatively small sample of 91 individuals used in this study was recruited from psychiatric clinics across the Oxfordshire and Buckinghamshire counties in England. While such patients are likely to be more representative than RCT participants, they may not be fully representative of the entire UK BD population. Second, we addressed the presence of missing data with multiple imputation of missing variables assuming a missing at random (MAR) mechanism to limit the loss of too many observations or variables for the analysis. Although there were no statistically significant differences between the baseline characteristics of the complete case patients (n = 48) and the overall sample (n = 91) and the complete case sensitivity analysis showed no significant impact of data gaps on the overall results, the lower mean costs estimated for the complete cases suggests that patients retained in the follow-up likely had better functioning outcomes. Third, due to the lack of data, productivity losses for unemployed individuals whose ongoing unemployment was attributed to BD and premature mortality due to BD were not included likely leading to some underestimation of the overall economic burden.

CONCLUSIONS
We estimate that the total annual cost-of-illness for BD amounted to attributed to premature mortality, the magnitude of costs reported in our present study mirrors the economic burden of schizophrenia (£6.7 billion) in the United Kingdom (Mangalore & Knapp, 2007). The immense costs of informal care provided by family and friends of BD patients, however, so far have received much less attention than in the case of schizophrenia and dementia. The identification of schizophrenia as the "paradigm condition" or the heartland of psychiatry and the relatively lower burden on health and social care has probably contributed to a relatively low priority for research into BD and resulting low research spend (Harrison et al., 2016).
The magnitude of health care and social care-related costs and the costs of informal care and due to lost productivity were significantly associated with self-reported levels of mood symptoms. From a public mental health perspective, these results generate the hypothesis that significant cost savings could result from improved monitoring and control of self-reported symptoms between episodes using a low-cost and relatively simple method of mood monitoring.
This study provides robust evidence that BD is a disease of substantial economic burden in the United Kingdom. Policy makers need to pay attention to this distribution of expenditure in BD as health care and medications represent a lesser proportion of the total economic burden, implying that the focus of any future interventions should be related to improving the functionality and productivity of individuals with BD.