Abstract
Background ‘High-cost’ individuals with multimorbidity account for a disproportionately large share of healthcare costs and are at most risk of poor quality of care and health outcomes.
Aim To compare high-cost with lower-cost individuals with multimorbidity and assess whether these populations can be clustered based on similar disease patterns.
Design and setting A cross-sectional study based on 2019/2020 electronic medical records from adults registered to primary care practices (n = 41) in a London borough.
Method Multimorbidity is defined as having ≥2 long-term conditions (LTCs). Primary care costs reflected consultations, which were costed based on provider and consultation types. High cost was defined as the top 20% of individuals in the cost distribution. Descriptive analyses identified combinations of 32 LTCs and their contribution to costs. Latent class analysis explored clustering patterns.
Results Of 386 238 individuals, 101 498 (26%) had multimorbidity. The high-cost group (n = 20 304) incurred 53% of total costs and had 6833 unique disease combinations, about three times the diversity of the lower-cost group (n = 81 194). The trio of anxiety, chronic pain, and depression represented the highest share of costs (5%). High-cost individuals were best grouped into five clusters, but no cluster was dominated by a single LTC combination. In three of five clusters, mental health conditions were the most prevalent.
Conclusion High-cost individuals with multimorbidity have extensive heterogeneity in LTCs, with no single LTC combination dominating their primary care costs. The frequent presence of mental health conditions in this population supports the need to enhance coordination of mental and physical health care to improve outcomes and reduce costs.
Introduction
Multimorbidity, the presence of ≥2 long-term conditions (LTCs), has become a major public health concern across healthcare systems.1–3 With an ageing population and higher prevalence of conditions, such as obesity in younger cohorts, multimorbidity is now the norm across Europe.4 In the UK, two-thirds of people aged >65 years are expected to live with multimorbidity by 2035.5 Primary care is organised into general practices, which are the first point of contact for health needs and play a major role in preventing, diagnosing, and caring for individuals with multimorbidity.6–8 Funding and recruitment shortfalls have resulted in increasing workload, with more patients per GP,9,10 challenging their ability to care properly for people with multimorbidity.11,12 This problem is magnified among ‘high-cost’ individuals with multimorbidity13 who have the highest risk of unmet health need, poorly coordinated and duplicated care, and worse health outcomes.14,15 Understanding the healthcare needs of these patients better can inform quality improvement and cost containment efforts.16
Little is known about high-cost individuals with multimorbidity. Research into the prevalence and combinations of LTCs has considered either the overall multimorbid population or separate age groups.17–20 Clusters of LTCs are defined by prevalence to date and may not correspond to the patient groups with worse outcomes and/or more frequent health service use.21 In a nationally representative sample from the UK, Zhu et al identified the cluster with a high prevalence of depression, anxiety, and pain as having the greatest health service use.18 Soley-Bori et al found that the disease/condition-based clusters of alcohol and substance dependence, followed by anxiety and depression, had the highest primary care use as additional LTCs develop over time.22 However, understanding the distribution and concentration of costs is needed,13 and identifying clusters based on primary care costs rather than disease prevalence may yield different results. International literature reports highly concentrated hospital costs, where the top 10% of patients account for between 50% and 80% of costs.23 Similar evidence in primary care, particularly within the multimorbid population in the UK, is lacking.
| ‘High-cost’ individuals with multimorbidity tend to have the highest risk of unmet health need, poorly coordinated and duplicated care, and worse health outcomes. Understanding the healthcare needs of these individuals is essential to inform quality improvement and cost containment efforts. To the authors’ knowledge, this is the first study to characterise high-cost individuals with multimorbidity based on their primary care costs and disease prevalence, and compare them with lower-cost individuals. The most expensive combinations of diseases were identified and whether high-cost individuals can be clustered based on similar patterns of disease was assessed using a young, urban, multi-ethnic population-based sample. |
This study aimed to:
characterise and compare high-cost with lower-cost individuals with multimorbidity in primary care, in terms of LTC combinations and their contribution to primary care costs; and
assess whether these populations can be grouped into clusters of similar LTCs.
Method
Study design, setting, participants, and data
This retrospective cross-sectional study of 386 238 electronic primary care medical records included adults (aged ≥18 years) registered between 1 April 2019 and 31 March 2020 at 41 general practices in the Lambeth DataNet. Individuals with multimorbidity (defined as the co-occurrence of ≥2 of 32 selected LTCs, Supplementary Box S1) totalled 101 498.
Patient-level primary care costs reflect primary care use and workload, measured by consultations. Consultations were costed for type of provider delivering care (GP, nurse, or other healthcare professional) and mode of delivery (face to face, telephone, home visits, or electronic). Other healthcare professionals included pharmacists, healthcare assistants/support workers, and physician associates. National unit costs from the year 2019–2020 were used for valuation, adjusted by average duration of consultation.24 For example, to cost a GP face-to-face consultation, the cost per hour for a GP (£255) was adjusted for the average duration of a GP face-to-face consultation (9.22 min), resulting in a unit cost of £39 (£255 × (9.22/60)).
High-cost individuals were defined as those at the top 20% (≥80th percentile) of the primary care cost distribution. Lower-cost individuals were the remaining sample (<80%). Although thresholds are often set at 10% or 5% quantiles,14,25 the current study’s more inclusive approach yields a larger sample size, still reflects higher-cost individuals, and allows assessment of the principle of cost disproportionality,26 where the relative medical expenditure of a population is larger than their relative size. Sensitivity analyses using the 70th and 90th percentile thresholds were conducted.
Statistical methods
The full distribution of primary care costs are described for high-and lower-cost groups. Demographic characteristics (age, sex, ethnicity, and Index of Multiple Deprivation [IMD 2019]), death rates, prevalence of LTCs, consultations, and costs were summarised. Differences in these variables between high- and lower-cost individuals were tested using t-tests (Mann–Whitney U-test for non-normal distributions) and Pearson χ2 tests (for categorical variables). Missing data were kept as missing (Supplementary Box S2).
With 32 LTCs, the possible number of disease combinations is extensive (496[ = 16 × 31] pairs, 4960[ = 32!/(29!×3!]) triads, 35 960 (= 32!/[28!×4!]) tetrads, for example). To characterise disease heterogeneity among high- and lower-cost individuals, LTC combinations were identified with average and total costs computed. This step identified combinations of LTCs with a major contribution to primary care costs.
Latent class analysis (LCA) explored whether multimorbid individuals could be grouped into a more parsimonious and interpretable set of clusters based on their underlying LTC prevalence. Each patient cluster captures a different LTC prevalence and combination, with up to five classes tested. The final model was chosen by balancing goodness of fit measure, model stability, and convergence.27 Entropy then measured model fit.28 The resulting clusters were described based on demographic characteristics, death rates, average consultations and costs, average number of LTCs, and prevalence of individual LTCs and LTC combinations. Further details are in Supplementary Box S2.
SAS (version 9.4) was used for analyses, including PROC LCA to implement the clustering technique. This study is reported using STROBE guidelines (https://www.strobe-statement.org).
Results
Population
High-cost individuals with multimorbidity (n = 20 304) were older, more often female, from ethnicities other than White, and more deprived than the lower-cost (n = 81 194) group (all P-values <0.0001) (Table 1). They had an average of 4.28 LTCs (standard deviation [SD] 1.99), with chronic pain, anxiety, hypertension, depression, and osteoarthritis the most prevalent LTCs.
Table 1. Description of study sample, individuals with multimorbidity in 2019/2020a
A higher prevalence of chronic pain (76.52% versus 50.19%), hypertension (45.74% versus 31.04%), and osteoarthritis (30.30% versus 15.61%) was observed among high-versus lower-cost individuals (P<0.0001). The high-cost group used primary care four times more than the lower-cost group (mean annual consultation 23.03 [SD 11.08] versus 5.74 [SD 4.66], P<0.0001).
Primary care costs and disease combinations
As consultation costs of high-cost individuals amounted to £13.1 million (Supplementary Table S1), this 20% of the sample incurred 53% of total consultation costs of the multimorbid population. Disproportionality starts at the 70th percentile, with costs for individuals in the 70th–80th percentile accounting for 14% of total costs (Supplementary Table S2).
The mean annual cost of consultations by high-cost individuals was £644 (SD 269) compared with £146 (SD 115) for lower-cost individuals. The cost distribution for high cost was more right-skewed and had a longer tail (Supplementary Table S1).
High-cost individuals had a total of 6833 unique combinations of LTCs, which represent 51% of all combinations in the whole sample (Supplementary Table S5) and nearly three times more heterogeneity than lower-cost individuals. The combination of LTCs with the largest contribution to costs was the trio of anxiety and chronic pain and depression (5.12%), followed by the pairs anxiety and chronic pain (2.52%), and anxiety and depression (2.37%) (Table 2). The top 10 combinations of LTCs, by share of costs, involve six LTCs: anxiety, chronic pain, depression, asthma, hypertension, and osteoarthritis. Sensitivity analyses showed similar results (Supplementary Tables S3 and S4).
Table 2. Top 10 LTC combinations based on their contribution to total primary care costs for high- cost individuals with multimorbidity, 2019/2020 (n = 20 304; total number of LTC combinations, n = 6833)
Among the lower-cost population, the combination of LTCs with the highest contributions to costs were also the pair anxiety and depression (6.92%), the trio anxiety and chronic pain and depression (5.87%), and the pair anxiety and chronic pain (3.89%) (Table 3). Osteoarthritis did not feature in the top 10 combinations for lower-cost individuals, whereas it did with the high-cost individuals, but diabetes did feature for the lower-cost individuals.
Table 3. Top 10 LTC combinations based on their contribution to total primary care costs, for lower-cost individuals with multimorbidity, 2019/2020 (n = 81 194; total number of LTC combinations, n = 9354)
LCA
Within high-cost individuals, five clusters proved the best grouping (Supplementary Table S6). No single combination of LTCs accounted for a large proportion of the clusters, except for cluster 1 where 15% of individuals had the trio anxiety and chronic pain and depression (Table 4). This cluster was younger (mean age 44 years, SD 14), with a high percentage of females (75%). For the remaining clusters, all LTC combinations had a prevalence <5%. In three of the five clusters, mental health conditions were prominent (30%, 63%, and 79% in clusters 1, 2, and 4 had depression, respectively), along with asthma (cluster 1), alcohol dependence (cluster 2), and chronic pain (cluster 4). In cluster 3, diabetes and osteoarthritis were the most prevalent LTCs (39% each). Individuals in cluster 5 showed a noticeable prevalence of cardiac diseases and chronic kidney disease, accompanied by other LTCs (average of 6.97 LTCs, SD 1.94), and had the highest average costs (£725, SD 336). They also had the highest average age (78.64 years, SD 11.37) and lowest proportion from deprived areas (24.47%).
Table 4. Results of LCA: clusters among high-cost individuals with multimorbidity, 2019/2020 (n = 20 304)
In four of the five clusters, the average count of LTCs was higher than the average of each of the lower-cost clusters. Clusters 2 and 4 had the highest rates of patients dying (3.58% and 3.23%, respectively); this was higher than any of the lower-cost clusters (Tables 4 and 5).
Table 5. Results of LCA: clusters among low-cost individuals with multimorbidity, 2019/2020 (n = 81 194)
Three clusters best grouped lower-cost individuals (Supplementary Table S7). All three were characterised by a relatively high prevalence of chronic pain (45%, 49%, and 46% across clusters 1, 2, and 3, respectively), along with diabetes (cluster 1), asthma (cluster 2), and depression (cluster 3) (Table 5). Unlike the high-cost group, each cluster had at least one combination of LTCs with >5% prevalence (for example, 6% in cluster 1 had diabetes and hypertension, and 23% in cluster 3 had anxiety and depression). Cluster 1 had the highest average cost at £173.09 (SD 112.19), the highest average age (65.43 years, SD 13.87), and lowest percentage of White ethnicity (43.04%).
Discussion
Summary
High-cost individuals with multimorbidity accounted for 53% of the total costs of primary care consultations. They were grouped into five clusters, with mental health (anxiety and depression) highly prevalent in three clusters. People with chronic pain were concentrated in one cluster, with a prevalence in the whole high-cost sample of 77%. Individuals with cardiac conditions dominated another cluster and alcohol dependence (along with depression and anxiety), asthma, and osteoarthritis were also present across clusters.
High-cost individuals are a more heterogeneous group of patients than those with lower costs. This is evidenced by: the higher mean number of LTCs (4.28 versus 3.03, P<0.0001), the larger number of clusters (five versus three), and the three times larger number of LTC combinations. High-cost individuals only showed combinations of LTCs >5% prevalence in one of five clusters, meaning no single combination of LTCs made a sizeable contribution to primary care costs, the largest being anxiety, chronic pain, and depression. This contrasts with the lower-cost group where each cluster had between one and five LTC combinations >5%, indicating a greater concentration in specific combinations of LTCs.
Strengths and limitations
This study advances existing literature on health service use and costs relating to individuals with multimorbidity by focusing on the patients with the most expensive cases. Descriptive analyses of combinations of LTCs and their contribution to costs are complemented by latent class analyses to understand clusters of individuals with similar patterns of LTCs. A large sample with urban, deprived, and ethnically diverse individuals was used, with sociodemographic, medical, and primary care use information. This fills an important research gap given the crucial role of primary care in managing an increasingly multimorbid population. The diverse population allows for meaningful analysis of ethnic minority groups, who are typically under-represented in health research.30 Costs reflect the workload of primary care providers. Identifying high-workload individuals is important given the current mismatch between workforce demand and supply.
Results may not be generalisable to rural, older, or less ethnically diverse populations. For example, having a smaller proportion of patients from ethnic minority groups may reduce prevalence of hypertension and diabetes,31,32 whereas older populations may have more multimorbidity given its correlation with age.33
Primary care costs reflect consultations only and the inclusion of medication costs may affect results since the cost of medication relative to total costs can vary by condition.34–36 Averages from the Personal Social Services Research Unit were used for consultation duration. Actual duration may vary by LTC. LTCs were defined based on diagnostic Read codes that may miss patients who do not have a formal diagnosis despite receiving treatment, the extent of which may vary by LTC.37 Similarly, the presence of a condition was defined as a diagnostic code before or during the study period and, therefore, any changes to diagnoses were not considered. Limited by cross-sectional data, the current study could not differentiate between 1-year high-cost individuals and people with more persistent conditions over time.25 The current study considers primary care only, and it is not clear whether conclusions would hold across secondary and social care settings, or the extent to which the degree of substitution and complementarity among care components should be accounted for.13 Finally, the current study describes costs only, and this debate should also be informed by outcomes.
Comparison with existing literature
Previous literature on clustering multimorbid individuals has not focused on the high-cost group, yet some similarities are apparent. Zhu et al identified the highest primary care use in clusters with depression, anxiety, and pain.18 These LTCs were highly prevalent among the clusters found in the current study. Soley-Bori et al pointed at the alcohol dependence and substance dependence cluster as the one with the highest expected primary care demand as further LTCs develop.22 In the current study, alcohol dependence was common in one of the clusters with high-cost individuals. Stokes et al concluded that no single combination of LTCs contributed significantly to total secondary care costs, with the pair diabetes and hypertension representing the highest share (3.2%).21 The current study reached similar conclusions based on primary care data. A total of 13 388 LTC combinations existed in the overall multimorbid sample, 6833 among the high-cost subgroup. No single combination made a major contribution to primary care costs, with anxiety, chronic pain, and depression accounting for just 5% of total primary care costs incurred by high-cost-multimorbid individuals.
Tran et al found the combination of cancer and mental health conditions was the most expensive.38 The current study did not find cancer to be a major contributor to primary care costs. However, Tran et al included hospital admissions and outpatient care, suggesting cancer incurs significant costs outside primary care compared with other LTCs.38
Implications for research and practice
Enhancing care coordination across specialties caring for individuals with multimorbidity is considered necessary for high-quality and efficient care.39 This strategy is particularly important for high-cost individuals, who use a disproportionate percentage of healthcare services. Understanding the most common clusters of LTCs may help to prioritise care coordination and integration efforts.2 Three of the five ‘high-cost’ clusters identified in this study had a high prevalence of anxiety and depression, suggesting mental health issues expand primary care use in this group. Prioritising mental health and enhancing its coordination with physical LTCs may improve outcomes and reduce costs among high-cost individuals. The Improving Access to Psychological Therapies for people with LTCs (called IAPT-LTC), currently underway, is an example of an initiative to facilitate access to mental health services and coordination between mental and physical health providers.40 Further research on the impact of mental health on healthcare needs, costs, and outcomes of individuals with multimorbidity is needed.
Findings from this study underscore the wide heterogeneity of high-cost individuals with multimorbidity and further research should consider if this is associated with increased clinical complexity and a higher risk of unmet need and/or poorer health outcomes. Individualised care, tailored and centred around each patient, reflecting their preferences — rather than one-size-fits-all strategies — seems like the appropriate response. This conflicts with limited primary care resources and mostly single-disease-oriented payment incentives such as the Quality and Outcomes Framework. To ease this tension, research is needed on enhanced payment mechanisms that transcend disease boundaries and explore the design of cost-effective interventions that facilitate — through technology, data, or other mechanisms — healthcare delivery for individuals with multimorbidity.
Acknowledgments
The process for selecting long-term conditions (LTCs) and risk factors for this study was led by Outcomes Based Healthcare (OBH), London, UK (https://outcomesbasedhealthcare.com). King’s College London and OBH devised a process to select Read, EMIS, and SNOMED codes for the presence of each of the included LTCs and risk factors; also, for the remission or relapse of LTCs and risk factors over the course of the study period.
Notes
Funding
This research was funded by
Impact on Urban Health, part of Guy’s & St Thomas’ Foundation, a charitable foundation (charity number: 1160316; grant number EIC180901). The funder did not contribute to the design of the study; the collection, analysis, and interpretation of data; or the writing of the manuscript. Mark Ashworth has been funded through UK Research and Innovation: ‘MELD-B’, NIHR203988; and the Medical Research Council: ‘Born in South London (eLIXIR)’, MR/X009742.
Ethical approval
All data were extracted under the terms of a signed data-sharing agreement with each practice and with project-specific approval following submission of a data privacy impact assessment, approved by Lambeth Clinical Commissioning Group on 2 November 2017. Information governance approval required ‘low number suppression’, ensuring that data could not be displayed if the patient number was ≤10 in any given category; in these circumstances, data reporting would state: ‘≤10 patients’. Separate ethical committee approval was not required (Health Research Authority, 29 September 2017) as all data were fully anonymised for the purposes of research access, and all patient-identifiable data had been removed.
Data
The datasets generated and/or analysed during the current study are not publicly available because the condition for pseudonymised data extraction by included general practices was that the data would not be shared beyond parties named in the data-sharing agreement, but are available from the corresponding author on reasonable request.
Provenance
Freely submitted; externally peer reviewed.
Competing interests
The authors have declared no competing interests.
