Sage Journals: Discover world-class research

Abstract

Objectives

To identify population groups with distinct multimorbidity patterns and to examine their associations with high treatment burden.

Methods

Latent class analysis was conducted using cross-sectional, self-reported data on 16 long-term conditions from Danish adults aged 25 years and older in treatment (N = 14,344), drawn from the Danish National Health Survey. Treatment burden was assessed using the Multimorbidity Treatment Burden Questionnaire, and associations between identified multimorbidity patterns and high treatment burden (score ≥22) were examined using a bias-adjusted three-step approach.

Results

Eight latent classes were identified: 1 Hypertension (31% of the study population); 2 Mental health disorders (21%); 3 Musculoskeletal disorders (17%); 4 Complex cardiometabolic-musculoskeletal disorders (10%); 5 Complex headache-mental-back-disorders (7%); 6 Asthma-allergy (6%); 7 Cataract-respiratory disorders (5%); and 8 Complex respiratory-musculoskeletal disorders (3%). Overall, 13% experienced a high treatment burden. This proportion varied from 0.5% in Class 3 to 48% in Class 5, with the highest proportions associated with Classes 4 (27%), 5 (48%), and 8 (26%). The three complex multimorbidity groups averaged more than four conditions per individual and, compared to Class 1, were associated with a non-Danish background, being temporarily or permanently out of work, low social support, and difficulties engaging with healthcare providers. The prevalence of high treatment burden and the strength of its association with disease patterns decreased with age, while no substantial differences were observed across sexes.

Conclusion

Specific multimorbidity patterns were strongly associated with high treatment burden. These findings may inform healthcare planning, resource allocation, and tailored interventions to reduce treatment burden.

Keywords

multimorbidity comorbidity cross-sectional survey latent class analysis public health

Introduction

The growing prevalence of multimorbidity poses significant challenges to patients’ well-being and healthcare systems worldwide. Handling these challenges is crucial for improving patient outcomes and ensuring integrated, well-coordinated, high-quality healthcare.^1–6 Individuals with multimorbidity face increased risks of reduced daily functioning, psychological distress, and mortality, which calls for integrated healthcare responses.^4,6–12 Yet, these patients frequently encounter fragmented and poorly coordinated services^1,4,13–16 with a high risk of overtreatment, contradictory treatments^14,16 and inappropriate healthcare utilisation.^8,10,17 Managing healthcare and treatment regimens in complex healthcare systems may often inflict a significant workload on patients alongside other daily responsibilities.^13,18 As a consequence, being in treatment may result in a significant patient-perceived treatment burden, defined as the effort required of patients to manage their health and its impact on daily life.¹⁹

The Cumulative Complexity Model underscores how imbalances between patient workload (e.g., scheduling and attending multiple appointments, managing various medications, navigating treatment instructions) and capacity (e.g., physical and mental functioning, financial resources, social support) contribute to healthcare inefficiencies and perceived treatment burden.²⁰ Treatment burden is increasingly recognised as a critical factor in understanding and supporting patients with multimorbidity,^21,22 as it can reduce compliance with treatment and health-related quality of life.²³ Consequently, strategies to improve care integration and health outcomes for patients with multimorbidity have attracted political, professional, and scientific attention. Yet significant gaps persist in our understanding of the patients’ needs, particularly concerning perceived treatment burden.

Previous research has shown that treatment burden is associated with specific long-term conditions, including diabetes,^24,25 mental illness,^24,26–28 and heart attack and stroke^24,28,29 as well as the overall number of conditions.^{24,25,27–29} However, simply counting conditions does not effectively capture the complexity of multimorbidity.^30–32 The common definition of multimorbidity (two or more concurrent diseases in an individual) identifies a large heterogeneous patient population with diverse needs, making it less useful for healthcare planning purposes. To address these challenges, researchers increasingly adopt statistical approaches to identify multimorbidity patterns.^33–35 Latent Class Analysis (LCA), a model-based approach for identifying distinct, unobserved categorical subgroups (latent classes) within a population based on observed indicators,³⁶ is the most dominant and promising of these approaches.^33–35 In multimorbidity research, LCA is beneficial as the number of possible disease combinations increases exponentially with the number of long-term conditions, making predefined disease groupings or simple disease counts impractical. By grouping individuals with similar patterns of observed disease indicators, LCA reduces complexity while potentially preserving meaningful disease patterns.

Despite numerous studies on multimorbidity patterns, their relationship with patient-perceived treatment burden has not yet been comprehensively examined. In our previous analysis of individuals with self-reported cardiovascular disease, we identified an increased risk of high treatment burden in individuals who also reported chronic obstructive pulmonary disease (COPD), musculoskeletal disorders, cancer, or mental disorders, compared to those with cardiovascular disease alone.²⁹ In contrast, the presence of diabetes alongside cardiovascular disease did not significantly increase the risk of a high treatment burden. Additionally, we found that individuals with low health literacy (difficulties in understanding health information) had a higher risk of high treatment burden. This burden increased markedly with the number of long-term conditions. These results underscore the importance of understanding how specific disease combinations affect treatment burden. Such knowledge could be critical for healthcare planning and reorganisation and may inform the design and implementation of interventions targeted at specific patient groups to reduce care fragmentation and improve health outcomes for patients with the most complex needs.

This study aimed to examine the associations between multimorbidity patterns and perceived treatment burden. First, we used LCA to group individuals in treatment from a large, cross-sectional survey into latent classes. Second, we characterised these classes using demographic, socioeconomic, and health-related factors. Third, we examined how the multimorbidity classes were associated with treatment burden.

Methods

Selection and description of participants

The Danish National Health Survey (DNHS) is a nationwide cross-sectional survey conducted every fourth year among Danish adult residents.^2,37 It employs stratified random sampling, drawing from the Danish Civil Registration System,³⁸ with one national subsample and five subsamples stratified by region and municipality. Of approximately 4.2 million Danish residents aged 25 years and older, 22.4% resided in the Central Denmark Region (CDR) in January 2021.³⁹ In total, 48,902 individuals (25+ years) from this region were invited to participate in the survey, of whom 62% (n = 30,307) responded to the questionnaire between February and May 2021, either using a secure web link or a paper questionnaire sent by post.

We included individuals aged 25 years and older from the 2021 CDR subsample who reported being in treatment and provided self-reported data on long-term conditions and perceived treatment burden. Being in treatment was defined as responding ”yes” to the question: ”Do you receive treatment or take medication for one or more conditions, or do you attend rehabilitation or regular check-ups?” (yes/no).²⁴ Only respondents self-reporting being in treatment (n = 14,549) were asked about their treatment burden.⁴⁰ Respondents with missing data on treatment burden and/or long-term conditions were excluded (n = 205), resulting in a final analytic sample of 14,344 individuals. A flow chart detailing participant inclusion is provided in Supplementary A.

The study was registered in the Central Denmark Region’s internal record of research projects (r. no. 1-16-02-307-22), and Central Denmark Region approved the use of the data. According to Danish law, no formal ethical approval is required for pseudonymised surveys and register-based research. All invited survey participants were informed that participation was voluntary, that their survey data could be used for research purposes, including potential linkage with administrative register data, and that full or partial survey completion constituted implied consent. All data were pseudonymised for analysis and reporting of study findings.

Data collection and measurements

Long-term conditions and multimorbidity

Information on long-term conditions was obtained using a modified version of a disease checklist recommended by the World Health Organization for health surveys.⁴¹ For each disease, respondents were asked whether they currently had, or previously had, the condition and continued to experience after-effects. The survey included 18 conditions selected based on their clinical significance and potential impact on daily life. Individuals with partial responses were retained as non-completed conditions were considered disconfirmed if at least one of the conditions on the list was completed.

To enhance data quality and model stability, reducing the risk of unstable and spurious class distinctions, we combined indicator variables that were conceptually similar and at risk of small cell counts if analysed separately. Specifically, angina pectoris and heart attack were grouped as “heart disease”, reflecting their shared ischaemic aetiology, overlapping risk factors, and similar management implications. Likewise, “mental illness lasting less than 6 months” and “mental illness lasting more than 6 months” were grouped as “mental illness” due to substantial overlap and limited empirical distinction. This resulted in a total of 16 distinct disease indicator variables (Table 1). Participants were classified as having multimorbidity if they reported two or more of the 16 included conditions.

Table 1.

Participant characteristics (participants from the 2021 DNHS, Central Denmark Region, 25+ years, self-reported in treatment, n=14,344).

	n	%^a
Sociodemographic factors
Mean age (SD), years	60	(16.3)
Age, years (missing: n=0; 0%)
25-34	797	10
35-44	1,167	10
45-54	2,138	16
55-64	3,342	21
65-74	3,882	23
75+	3,018	20
Sex (missing: n=0; 0%)
Male	6,476	46
Female	7,868	54
Country of origin (missing: n=0; 0%)
Danish	13,585	92
Non-Danish	759	8
Educational level (missing: n=514; 3.6%)
Low (0-10 years)	2,508	20
Medium (11-14 years)	7,345	52
High (15+ years)	3,977	28
Employment status (missing: n=172; 1.2%)
Employed or enrolled in education	5,966	44
Unemployed	902	8
Permanently out of work	7,304	48
Cohabitation status (missing: n=214; 1.5%)
Living with spouse/partner	10,297	67
Not living with spouse/partner	3,833	33
Children (<16 years) in the household (missing: n=1,149; 8.0%)
Yes	1,944	17
No	11,251	83
Social support (missing: n=302; 2.1%)
High	11,804	83
Low	2,238	17
Understanding health information (missing: n=410; 2.9%)
Easy	13,249	95
Difficult	685	5
Actively engage with healthcare providers (missing: n=407; 2.8%)
Easy	12,875	92
Difficult	1,062	8
Self-rated health (missing: n=158; 1.1%)
Good	10,381	72
Poor	3,805	28
Self-reported long-term conditions (missing: n=0; 0%)
Hypertension	6,055	39
Heart disease^b	855	6
Stroke	615	4
Diabetes	1,696	12
Cancer	1,140	7
COPD	1,417	10
Asthma	1,777	13
Allergy	3,037	23
Osteoarthritis	5,375	35
Rheumatoid arthritis	1,758	12
Osteoporosis	1,371	9
Back disorders	3,393	23
Mental illness	2,196	18
Migraine or recurrent headache	2,706	20
Tinnitus	3,069	21
Cataract	1,344	9
Number of long-term conditions (missing: n=0; 0%)
0	1,217	9
1	3,015	21
2-3	6,233	43
4+	3,879	27
Mean number of long-term conditions (SD)	2.62	(1.8)
Treatment burden (missing: n=0; 0%)
None (score 0)	5,466	36
Low (score >0 and <10)	4,591	31
Medium (score ≥10 and <22)	2,653	20
High (score ≥22)	1,634	13
Mean treatment burden score (SD)	9.11	(13.5)

^aWeighted to represent the population of the Central Denmark Region, aged 25+ years, in treatment.

^bIncludes angina pectoris and heart attack.

DNHS = Danish National Health Survey; SD = standard deviation; COPD = Chronic obstructive pulmonary disease.

Treatment burden

Patient-perceived treatment burden was measured using the Danish version of the Multimorbidity Treatment Burden Questionnaire (MTBQ).²⁴ The MTBQ is a ten-item validated measure that captures key aspects of treatment burden, including medication management, healthcare coordination and attendance, and self-monitoring.^27,42 Participants rated the difficulty of each item on a five-point Likert scale from 0 (not difficult/does not apply) to 4 (extremely difficult). We excluded participants with missing responses to more than 50% of MTBQ items. For the remaining participants, the mean score of answered items was multiplied by 25 to generate a global score ranging from 0 to 100.²⁴ We categorised scores as no burden (0), low burden (>0 and <10), medium burden (≥10 and <22), and high burden (≥22) and used a binary variable for high treatment burden (≥22) versus not high (<22) in the analyses.^24,27

Descriptive factors

To characterise study participants and identified disease classes, we included demographic, socioeconomic, and health-related characteristics: age, sex (male/female), and country of origin (Danish/non-Danish) from administrative register data linked to the survey data using a unique personal identification number assigned to all Danish citizens³⁸; self-reported educational level (classified using the Danish version of the International Standard Classification of Education⁴³ as low [0-10 years], medium [11-14 years], and high [15+ years]), employment status (employed or enrolled in education, unemployed, or permanently out of work [disability pension, early retirement pension, old age pension]), cohabitation status (living with a spouse/partner or not), presence of children in the household (living with child(ren) aged 0-15 years or not), social support, health literacy, and self-rated health from the DNHS. Perceived social support was measured using the single-item: “Do you have anyone to talk to if you have problems or need support?” with four response options, which we dichotomised as low (“Yes, sometimes” and ”No, never or almost never”) and high (”Yes, always” and ”Yes, mostly”).⁴⁴ Health literacy was assessed using two subscales from the Health Literacy Questionnaire^45,46: “Understanding health information well enough to know what to do” and “Ability to actively engage with healthcare providers”, each consisting of five items measured on a four-point Likert scale from 1 (”Very difficult”) to 4 (”Very easy”). We calculated each scale score as the mean of the five-item scores and standardised them to range from 1 (lowest ability) to 4 (highest ability). We then dichotomised each scale into difficult (score ≤ 2) and easy (score > 2) to dentify individuals who found it difficult compared with those who found it easy to understand health information or to actively engage with healthcare providers.²⁹ Self-rated health was measured using a single item and was dichotomised as poor (response options ”fair”' or ”poor”) and good (response options “excellent”, “very good” or “good”).

Statistics

Statistical analyses were conducted in Latent GOLD version 6.0. Stata version 18.5 and RStudio were used for graphical representations. We applied survey weights in the analyses to correct potential biases due to unequal selection probabilities and nonresponse.³⁷ These model-based weights, constructed by Statistics Denmark, increase representativeness by calibrating the data to known population characteristics based on register data.⁴⁷

Step 1: Identification of multimorbidity patterns

We used LCA to group individuals into latent classes with similar disease patterns. We employed an exploratory approach, estimating models with an increasing number of latent classes (K=1–15) without including covariates.^48,49 To guide model selection, we evaluated the relative model fit indices (Akaike Information Criterion [AIC], consistent AIC [CAIC], Bayes Information Criterion [BIC], and sample-size adjusted BIC [SABIC]) in addition to the log-likelihood values and clinical judgement to balance statistical fit with parsimony and substantive interpretation.³⁶ Lower fit indices and log-likelihood values indicate a better fit.³⁶ We emphasised the BIC, which is robust for LCA in large samples.⁵⁰ As standard likelihood ratio tests are inaccurate for model selection in LCA, we used the Vuong-Lo-Mendell-Rubin (VLMR) adjusted likelihood ratio test to evaluate model improvement with an increasing number of classes.⁵⁰ Among the best candidate models, we inspected the standardized residuals, comparing observed to estimated response pattern frequencies to evaluate the absolute fit of the models to the data (|std.res| > 3 indicating poorly fit response patterns).⁵¹

We subsequently used classification diagnostics to evaluate the quality of the preferred model. Both within-class homogeneity and between-class separation are essential for reliable interpretation when a model is used in a multistep approach.⁵¹ LCA relies on the assumption that the K latent classes explain the associations between the observed indicators, thereby causing the indicators to be statistically independent within classes (local independence assumption), meaning that the class-assigned individuals have relatively homogeneous response patterns. We used the bivariate residuals (BVRs, satisfactory thresholds <4) to evaluate whether this assumption was fulfilled. To assess the across-class separation, meaning how well the classes are differentiated, we used the entropy R² (range 0-1), with values ≥0.8 typically considered acceptable.³⁶ However, lower entropy values are common in exploratory LCA.⁵² Class sizes were also judged to avoid minimal solutions.

The primary outcomes of the final chosen model were the estimated class sizes (proportion of individuals assigned to each class; range 0-1), the item-response probabilities (probability of each long-term condition within each class; range 0-1), and K posterior probabilities (range 0-1) providing the probability for each individual of belonging to each of the K classes.

Step 2: Characterisation of multimorbidity patterns

Using bivariate analyses, we characterised the identified multimorbidity classes based on the demographic, socioeconomic, and health-related factors included in the study. Additionally, we examined how these factors predicted class membership using a bias-adjusted three-step latent class (LC) approach with covariates.^52–54 This method estimates multinomial logistic regression coefficients and standard errors for covariate effects on class membership while adjusting for classification uncertainty in the assignment of individuals to the latent classes. Odds ratios (ORs) and 95% confidence intervals (CIs) were calculated based on the model output.

Step 3: Association between multimorbidity patterns and high treatment burden

We examined the association between multimorbidity patterns and high treatment burden using a bias-adjusted three-step LC approach with a binary distal outcome. This approach employs binary logistic regression to estimate the effect of class membership on the outcome while adjusting for classification uncertainty. The model provides class-specific estimated probabilities and standard errors for high treatment burden, from which 95% CIs were calculated. Overall and pairwise differences between disease classes were evaluated using Wald tests.

As recommended when estimating associations between latent classes and covariates or categorical outcomes, we used proportional assignment and maximum likelihood (ML) adjustments with robust standard errors.^49,52,54,55 Missing values on covariates were handled using the mean imputation procedure available in Latent GOLD.⁵⁶

To evaluate the robustness of the findings, we first conducted supplementary analyses using respondents from the 2017 CDR subsample.⁵⁷ We then repeated the analysis of class-specific probabilities of high treatment burden restricted to individuals with a posterior probability >0.7. Furthermore, we employed two alternative MTBQ cut-offs to define high treatment burden (≥20 and ≥25). Finally, to examine whether associations between the identified multimorbidity patterns and high treatment burden varied in strength or direction by demographic subgroups, we conducted supplementary analyses stratified by age and sex, respectively, using the bias-adjusted three-step LC approach described above.

Results

Participant characteristics

The study included 14,344 participants aged 25 years and older who self-reported being in treatment (Table 1). The mean age was 60 years (standard deviation [SD] 16.3), and 54% were female. Almost half (48%) were permanently out of work, and 28% had poor self-rated health. Among the 16 long-term conditions, the prevalence ranged from 39% (hypertension) to 4% (stroke), and 11 conditions had a prevalence of 10% or more. Most participants (70%) had multimorbidity, and the mean number of long-term conditions was 2.62 (SD 1.8). The mean treatment burden score was 9.11 (SD 13.5), and 13% experienced a high treatment burden (≥22).

Compared to the total Central Denmark Region subsample aged 25 years and older of the 2021 DNHS, the main sample included a higher proportion of older adults and, within each of three stratified age groups, consistently poorer health (see Supplementary B).

The 16 conditions yielded a total of 65,536 possible response patterns (disease combinations), of which 2,618 were observed in the data, with 86% having very low frequencies with fewer than five participants exhibiting each of these patterns. A detailed list of the most frequent response patterns, along with their mean age and mean treatment burden, are presented in Supplementary C.

Selection of the latent class model

We reviewed the models with one to 11 latent classes as the models with 12 to 15 classes were not well identified (Table 2). The BIC favoured the eight- and nine-class models (differed by decimals). The CAIC also favoured the eight-class model, while the SABIC favoured the 10-class model, and AIC did not reach a minimum level. The VLMR test indicated a statistically better model fit for each extra class included. An elbow plot of the fit indices, presented in Supplementary D, indicates that the relative improvement in model fit was low after including six or seven classes (except for AIC). Following a detailed inspection of the models with six to nine classes, including standardized residuals and the interpretation of the models, we opted for the eight-class model. The standardised residuals indicated that 20% of observed disease patterns were poorly fit (Supplementary D). However, only 2% of disease patterns with estimated frequencies greater than 1 showed poor fit, suggesting that the model adequately represented the most prevalent patterns.

Table 2.

Fit statistics and diagnostic criteria for the estimated latent classes.

Number of LCs	npar	LL	BIC	SABIC	AIC	CAIC	VLMR^a	Max BVR	Entropy	Classification error	Smallest class size (%)^b
1	16	-82,832	165,814	165,763	165,695	165,830		1,216.2	1.00	0.00	100.0
2	33	-81,038	162,387	162,282	162,142	162,420	3,587	1,111.7	0.55	0.12	32.5
3	50	-80,131	160,735	160,576	160,363	160,785	1,813	455.3	0.52	0.23	26.2
4	67	-79,736	160,105	159,892	159,606	160,172	791	151.9	0.59	0.22	8.6
5	84	-79,518	159,829	159,562	159,204	159,913	436	145.1	0.58	0.25	7.0
6	101	-79,356	159,666	159,345	158,915	159,767	324	85.8	0.58	0.27	6.0
7	118	-79,208	159,530	159,155	158,652	159,648	296	45.5	0.56	0.30	4.4
8	135	-79,110	159,495	159,066	158,491	159,630	195	18.4	0.56	0.31	3.0
9	152	-79,030	159,495	159,012	158,364	159,647	161	15.1	0.56	0.32	2.5
10	169	-78,969	159,534	158,997	158,277	159,703	121	12.9	0.59	0.32	2.4
11	186	-78,922	159,601	159,009	158,217	159,787	94	13.1	0.60	0.32	2.1
12-15	Not well identified

^aAll p-values < 0.0001.

^bWeighted to represent the population of the Central Denmark Region 2021, aged 25+ years, in treatment. Bold values indicate the minimum value for each fit indices (BIC, ABIC, AIC, CAIC). LCs = Latent classes; npar = Number of estimated parameters; LL = Log-Likelihood; BIC = Bayesian Information Criterion; SABIC = Sample-size Adjusted BIC; AIC = Akaike Information Criterion; CAIC = Consistent Akaike Information Criterion; VLMR = Vuong-Lo-Mendell-Rubin adjusted likelihood ratio test; Max BVR = the largest bivariate residual between the 16 indicators.

Based on the diagnostic criteria, the model did not adequately capture all local dependencies as relatively large BVRs remained, the largest being among hypertension and diabetes (BVR 18.4, c.f. Supplementary E). In total, 28 (23%) of 120 BVRs were >4. The model had a moderate ability to allocate individuals into the eight classes (entropy = 0.56), indicating that a subset of individuals was difficult to assign to one specific class. According to the classification error, 31% of the individuals were misclassified. A detailed summary of misclassification proportions between the eight classes is presented in Supplementary E. The average posterior probability after modal assignment, i.e., after individuals were assigned to the class for which they had the highest posterior probability, served as a further indicator of assignment certainty. It was highest in Class 6 (0.81), followed by Class 8 (0.75) and Class 1 (0.74). Class 3 had the lowest average probability (0.61), while the remaining classes ranged from 0.65 to 0.71 (Classes 2, 5, and 7: 0.65/0.66; Class 4: 0.71).

Latent class disease profiles

Table 3 presents the class sizes, class-specific probabilities of having any of the 16 conditions, and the assigned labels for each class based on signature conditions in their disease profiles (i.e., conditions with high class-specific probabilities). Due to moderate class separation and misclassifications, we additionally used the disease profiles of individuals with posterior probabilities ≥0.8 after modal assignment to indicate each class’s signature conditions. Class-specific graphs illustrating smoothed disease profiles across posterior probability ranges are presented in Supplementary F.

• Class 1 Hypertension (31% of the study population): Characterised by a high probability of hypertension (59%) and a relatively high probability of diabetes (20%) compared to several other classes.

• Class 2 Mental health disorders (21%): Did not have a distinct signature condition but had a low to relatively low probability across all 16 conditions. The highest probability was for mental illness (34%), which, while not typical of a signature condition, was the second-highest across all classes.

• Class 3 Musculoskeletal disorders (17%) and Class 4 Complex cardiometabolic-musculoskeletal disorders (10%): Both had elevated risks of musculoskeletal disorders, but Class 4 also exhibited the highest risk of cardiovascular diseases and severe multimorbidity (99% with ≥2 conditions, mean = 4.8 conditions per individual).

• Class 5 Complex headache-mental-back disorders (7%): Had the highest risk of four of the 16 conditions, 97% with multimorbidity, and an average of 4.12 conditions per individual.

• Class 6 Asthma-allergy (6%): Characterised by the highest probabilities of asthma and allergy across all classes.

• Class 7 Cataract-respiratory disorders (5%): Defined by elevated risks of several conditions, with COPD as the signature condition and a notably increased probability of cataract compared to other classes.

• Class 8 Complex respiratory-musculoskeletal disorders (3%): Exhibited severe multimorbidity, with 100% having ≥2 conditions and a mean of 6.53 conditions per individual.

Table 3.

Latent class profiles (class proportions, class-specific disease probabilities) in the eight-class model.

Item-response probabilities > 0.5 in bold to assist interpretation. Underscored values represent the highest response probability within each item (disease indicator variable). COPD = Chronic obstructive pulmonary disease; Rheu. = Rheumatoid; Migraine = Migraine or recurrent headache.

Table 3 summarises the class-specific proportion of individuals with multimorbidity, the mean number of conditions per individual, and the proportion with poor self-rated health, using descriptive output unadjusted for classification uncertainty. Multimorbidity was most prevalent in Classes 3 to 8 (80-100%), with the mean number of conditions ranging from 2.64 in Class 3 (Musculoskeletal disorders) to 6.53 in Class 8 (Complex respiratory-musculoskeletal disorders). In contrast, multimorbidity was least prevalent in Classes 1 (Hypertension; 58% ≥2 conditions, mean = 1.86) and 2 (Mental health disorders; 41% ≥2 conditions, mean = 1.34). Poor self-rated health varied from 18% in Class 1 (Hypertension) to 56% in Class 8 (Complex respiratory-musculoskeletal disorders).

Demographic, socioeconomic and health-related characterisation of the latent classes

The disease classes differed in their sociodemographic composition, with notably older participants in Classes 4 (Complex cardiometabolic-musculoskeletal disorders) and 7 (Cataract-respiratory disorders) (>75% aged 65 years and older) and predominantly younger participants in Classes 2 (Mental health disorders) and 6 (Asthma-allergy) (>50% aged under 45 years). Classes 2 and 6 also had the highest proportions of highly educated individuals, whereas Classes 4, 7, and 8 (Complex cardiometabolic-musculoskeletal, Cataract-respiratory, and Complex respiratory-musculoskeletal disorders, respectively) had the largest shares with low educational attainment. Males were more prevalent in Class 1 (Hypertension), while females dominated most other classes except Class 7 (Cataract-respiratory disorders). See Supplementary G for details.

Since the study sample consisted of individuals in treatment, no distinct “Relatively healthy” class emerged. Class 1 (Hypertension) was therefore chosen as the reference category, as it represented the largest proportion of the sample and had a relatively lower burden of multimorbidity compared to most other classes. Against this reference, sociodemographic characteristics were significantly associated with latent disease class membership (ORs and 95% CIs shown in Table 4). Older age increased the likelihood of belonging to Class 4 (Complex cardiometabolic-musculoskeletal disorders) and Class 7 (Cataract-respiratory disorders) but decreased the likelihood of membership in the remaining classes, most notably Classes 2, 5, and 6 (Mental health, Complex headache-mental-back, and Asthma-allergy disorders, respectively). Females were more likely to be in all classes, with the strongest associations in Classes 2, 3, and 5 (Mental health, Musculoskeletal, and Complex headache-mental-back disorders, respectively). Lower educational levels were associated with higher odds of membership in Classes 4, 7, and 8 (Complex cardiometabolic-musculoskeletal, Cataract-respiratory, and Complex respiratory-musculoskeletal disorders, respectively), whereas higher education increased the likelihood of membership in Classes 2, 3, and 6 (Mental health, Musculoskeletal, and Asthma-allergy disorders, respectively). Being unemployed or permanently out of work was strongly associated with Classes 5 (Complex headache-mental-back disorders) and 8 (Complex respiratory-musculoskeletal disorders). Individuals not living with a spouse/partner and those with low social support were more likely to belong to Classes 4 (Complex cardiometabolic-musculoskeletal disorders) and 8 (Complex respiratory-musculoskeletal disorders), while low social support showed the strongest association with Class 5 (Complex headache-mental-back disorders). Difficulties in actively engaging in healthcare increased the likelihood of membership in all classes, with the most pronounced associations in Classes 5 (Complex headache-mental-back disorders) and 7 (Cataract-respiratory disorders). Full details are presented in Table 4.

Table 4.

Multinominal logistic regression results for covariate effects on disease class membership.

Latent	Class 1	Class 2		Class 3		Class 4		Class 5		Class 6		Class 7		Class 8		Wald test p-value
Assigned label	Hyper-tension (Ref.)	Mental health disorders		Musculoskeletal disorders		Complex cardio-metabolic-musculoskeletal disorders		Complex headache-mental-back-disorders		Asthma-allergy		Cataract-respiratory disorders		Complex respiratory-musculoskeletal disorders
Characteristics	OR	OR	95% CI	OR	95% CI	OR	95% CI	OR	95% CI	OR	95% CI	OR	95% CI	OR	95% CI
Age (per 5-year increase)	1	0.36	(0.33-0.39)	0.93	(0.87-0.98)	1.12	(1.06-1.19)	0.49	(0.46-0.53)	0.38	(0.35-0.42)	1.18	(1.08-1.29)	0.74	(0.68-0.81)	p < 0.0001
Sex
Male (ref.)	1	1		1		1		1		1		1		1		p < 0.0001
Female	1	3.79	(2.98-4.81)	4.09	(3.31-5.05)	1.62	(1.34-1.96)	5.31	(4.02-7.03)	2.71	(2.12-3.47)	1.49	(1.13-1.96)	2.80	(2.09-3.76)
Country of origin
Danish (ref.)	1	1		1		1		1		1		1		1		p < 0.0001
Non-Danish	1	1.16	(0.74-1.81)	0.73	(0.44-1.21)	1.61	(1.13-2.29)	1.48	(0.98-2.25)	0.91	(0.56-1.50)	1.01	(0.50-2.05)	2.54	(1.65-3.92)
Educational level
High (15+ years) (ref.)	1	1		1		1		1		1		1		1		p < 0.0001
Medium (11-14 years)	1	0.51	(0.39-0.67)	0.78	(0.63-0.97)	1.46	(1.11-1.92)	1.16	(0.84-1.60)	0.40	(0.30-0.53)	1.66	(1.12-2.46)	1.33	(0.89-1.98)
Low (0-10 years)	1	0.45	(0.31-0.64)	0.62	(0.46-0.83)	1.91	(1.41-2.59)	0.84	(0.56-1.27)	0.37	(0.25-0.55)	1.85	(1.21-2.81)	1.71	(1.12-2.61)
Employment status
Employed (ref.)	1	1		1		1		1		1		1		1		p < 0.0001
Unemployed	1	1.09	(0.73-1.61)	0.79	(0.47-1.31)	1.59	(0.86-2.93)	3.24	(2.16-4.87)	0.76	(0.48-1.20)	1.62	(0.69-3.81)	2.90	(1.68-5.02)
Permanently out of work	1	0.86	(0.63-1.18)	1.23	(0.95-1.60)	2.29	(1.60-3.26)	3.35	(2.40-4.66)	1.26	(0.90-1.77)	1.97	(1.13-3.44)	4.33	(2.67-7.03)
Living with spouse/partner
Yes (ref.)	1	1		1		1		1		1		1		1		0.002
No	1	0.97	(0.75-1.24)	1.03	(0.83-1.27)	1.43	(1.17-1.75)	1.08	(0.83-1.41)	0.77	(0.59-1.01)	1.10	(0.82-1.47)	1.33	(1.01-1.76)
Children in the household
No (ref.)	1	1		1		1		1		1		1		1		0.016
Yes	1	0.95	(0.70-1.31)	0.67	(0.43-1.04)	0.92	(0.57-1.47)	1.04	(0.73-1.49)	0.70	(0.51-0.97)	0.54	(0.28-1.06)	0.78	(0.46-1.33)
Social support
High (ref.)	1	1		1		1		1		1		1		1		p < 0.0001
Low	1	1.19	(0.89-1.59)	1.06	(0.78-1.44)	1.62	(1.26-2.09)	2.84	(2.14-3.76)	1.24	(0.91-1.70)	1.19	(0.81-1.73)	2.31	(1.67-3.19)
Understand
Easy (ref.)	1	1		1		1		1		1		1		1		0.160
Difficult	1	0.84	(0.48-1.47)	0.62	(0.34-1.14)	1.19	(0.78-1.81)	1.00	(0.61-1.65)	0.78	(0.40-1.52)	0.44	(0.22-0.86)	0.93	(0.45-1.92)
Actively engage
Easy (ref.)	1	1		1		1		1		1		1		1		p < 0.0001
Difficult	1	1.83	(1.22-2.75)	1.10	(0.65-1.86)	1.96	(1.26-3.05)	3.30	(2.20-4.95)	1.66	(1.01-2.71)	2.75	(1.62-4.67)	2.03	(1.08-3.82)

Results from multinomial logistic regression estimating the associations between covariates and disease class membership (N = 14,344), using Class 1 (Hypertension) as the reference category. Odds ratios (OR) and 95% confidence intervals (95% CI) are shown. Each OR is adjusted for all other variables in the model. Estimates account for classification uncertainty using a three-step latent class (LC) model with maximum likelihood (ML) correction and proportional assignment. Wald test p-values indicate the overall significance of each covariate across all latent classes. OR = Odds ratio; 95% CI = 95% confidence interval; Employed = Employed or enrolled in education; Understand = Understanding health information; Actively engage = Actively engage with healthcare providers.

Associations between latent classes and perceived treatment burden

Figure 1 shows significant variations in the estimated probability of high treatment burden across the distinct disease groups, ranging from 0.5% in Class 3 (Musculoskeletal disorders) to 47.8% in Class 5 (Complex headache-mental-back-disorders) (Wald test for overall difference across classes: 423.516, p < 0.0001). Compared to the overall probability of high treatment burden (13%), the probability was below average in Classes 1 (Hypertension) and 3 (Musculoskeletal disorders) and above average in Classes 4, 5, and 8 (Complex cardiometabolic-musculoskeletal, Complex headache-mental-back, and Complex respiratory-musculoskeletal disorders, respectively). While Class 5 (Complex headache-mental-back disorders) exhibited the highest proportion of individuals with high treatment burden (47.8%), Classes 4 (Complex cardiometabolic-musculoskeletal disorders) and 8 (Complex respiratory-musculoskeletal disorders) had similar levels (26.2-27.1%). Likewise, the proportion of individuals experiencing high treatment burden was comparable in Classes 2, 6, and 7 (Mental health disorders: 15.1%, Asthma-allergy: 12.5%, and Cataract-respiratory disorders: 15.1%, respectively). Details on the Wald tests for pairwise comparisons are presented in Supplementary H.1. Supplementary analyses on the 2017 sample support the findings.

Figure 1.

Associations between latent disease classes and high treatment burden in the eight-class model. Class-specific probabilities with 95% confidence intervals (CI) for high treatment burden were estimated using a three-step latent class (LC) model with maximum likelihood (ML) correction and proportional assignment (N = 14,344). The vertical line represents the overall prevalence of high treatment burden in the entire sample (13%), with 95% CI shown. Class-specific probabilities in % are shown in brackets. MTBQ = Multimorbidity Treatment Burden Questionnaire.

The robustness check using individuals with a posterior probability >0.7 resulted in a pattern consistent with the main findings (see Supplementary H.2). Sensitivity analyses applying alternative MTBQ cut-offs (≥20 and ≥25) likewise produced very similar class-specific patterns, although the absolute prevalence of high treatment burden varied slightly as expected (see Supplementary H.3).

Stratified analyses

The proportion of individuals experiencing high treatment burden and the strength of its association with disease patterns decreased with age, while the direction of the association remained relatively consistent across age groups. In contrast, no substantial differences were estimated between males and females (see Supplementary H.4).

Discussion

Using latent class analysis (LCA), we identified eight distinct disease classes among Danish adults aged 25 years and older who self-reported being in treatment, with each class representing between 3% and 31% of the sample. Compared with the overall prevalence of high treatment burden (13%), three classes were associated with particularly high proportions: Class 4 Complex cardiometabolic-musculoskeletal disorders (27%), Class 5 Complex headache-mental-back disorders (48%), and Class 8 Complex respiratory-musculoskeletal disorders (26%). Each of these three complex multimorbidity groups had an average of more than four conditions per individual, and together they comprised 20% of individuals aged 25 years and older in treatment, corresponding to almost one in ten individuals in the total population within this age group. Patterns of association were consistent in stratified analyses, with decreasing strength of associations between class membership and high treatment burden with increasing age, and no substantial sex differences despite variation in class composition.

Our study also indicated sociodemographic inequalities in the odds of belonging to the three complex multimorbidity groups with high treatment burden. Compared to Class 1 (Hypertension), which had less severe multimorbidity and a lower probability of high treatment burden, the odds of belonging to the complex multimorbidity groups were associated with a non-Danish background, being temporarily or permanently out of work, lacking social support, and having difficulties engaging with healthcare providers. Additionally, individuals with low educational levels and those living without a partner had higher odds of belonging to Class 4 (Complex cardiometabolic-musculoskeletal disorders) and Class 8 (Complex respiratory-musculoskeletal disorders) than Class 1 (Hypertension). This suggests an overrepresentation of an imbalance between patient workload and patient capacity in these three groups, consistent with the cumulative complexity model. However, given the cross-sectional design, these observed associations between sociodemographic factors and class membership may reflect both potential causes and consequences of multimorbidity and treatment burden, and the direction of these relationships cannot be determined from this study.

Our population-based study both supports well-established associations – that multimorbidity is common among those in treatment, that treatment burden tends to increase with the number of conditions, and that sociodemographic inequalities exist – and adds nuance by demonstrating substantial heterogeneity within these general patterns. First, multimorbidity profiles among individuals in treatment were highly diverse. Second, disease profiles with similar condition counts differed in poor self-rated health. They also varied markedly in treatment burden, indicating the added clinical relevance of pattern-based classification. Specific constellations – particularly those combining a high likelihood of mental health disorders and pain-related conditions such as osteoarthritis, back disorders, and migraine – appeared especially burdensome. Third, a notable share of working-age adults were represented in the high-burden groups. Finally, stratified analyses provided new evidence, showing that relative differences in the prevalence of high treatment burden between multimorbidity patterns were broadly consistent across sexes and age groups. However, the strength of the associations attenuated with age.

Our findings provide valuable guidance for clinicians in identifying patient groups at elevated risk of high treatment burden who could benefit from targeted interventions to reduce treatment complexity. Patients with complex multimorbidity involving, for example, cardiometabolic, respiratory, musculoskeletal, headache, or mental health disorders could be flagged in electronic medical records to increase clinician awareness. These groups might particularly benefit from targeted interventions aimed at reducing patient workload through, e.g., adjusted therapy intensity and improved care coordination, and strengthening patient capacity through structured social support initiatives like, e.g., patient navigator programs. Moreover, our results highlight the importance of addressing treatment burden not only among older patients with severe multimorbidity but also within working-age populations.

From a policy perspective, the identification of high-risk multimorbidity patterns may inform resource allocation and service planning, support integrated care pathways across sectors, and provide a framework for targeting and evaluating interventions in those most likely to benefit.

The identified multimorbidity patterns dominated by cardiometabolic, musculoskeletal, respiratory, or mental health disorders are consistent with existing literature.^12,34,58 However, unlike general population studies, our study exclusively included individuals in treatment, resulting in no distinguishable ‘Relatively healthy’ group. Consequently, despite higher disease prevalence than a healthy group the largest class, Class 1 (Hypertension), was a pragmatic reference group. Nonetheless, we identified significant sociodemographic associations with class membership relatable to previous findings of age distributions in disease prevalence and overrepresentation of females in all multimorbidity groups, except the ‘Hypertension’ group, especially those dominated by musculoskeletal, headache, and mental health disorders. Differences in healthcare-seeking behaviour and contact frequency between men and women may contribute to the observed sex differences in class membership.

Additionally, our results support existing evidence that individuals with complex multimorbidity patterns are characterised by lower social position, lower social support, and difficulties in actively engaging with healthcare providers. Furthermore, the finding of strong associations between complex multimorbidity patterns dominated by mental health disorders or cardiovascular diseases and high treatment burden support previous research showing an elevated risk of treatment burden related to these specific conditions.^24–29 Our findings also complement earlier results in a population with cardiometabolic disorders, which highlighted mental health disorders, musculoskeletal disorders, and COPD as significantly contributing to the perceived treatment burden.²⁹

Methodological strengths and limitations

The key strengths of this study include the large, population-based survey with high response rates, the use of a validated treatment burden measure (MTBQ), and the application of bias-adjusted three-step latent class models accounting explicitly for misclassification. Although moderate entropy suggested classification uncertainty, previous research supports the robustness of the three-step approach in large samples (n>10,000).^52,54 Calibrated weighting further enhanced the representativeness of the findings.

Several limitations deserve attention. First, self-reporting of ‘being in treatment’ may lead to misclassification if respondents misunderstood the question or included conditions beyond those surveyed. Indeed, 1,217 respondents reported being in treatment despite indicating none of the surveyed conditions, suggesting conditions outside the predefined list. This definition may also theoretically include a small group of individuals who attend regular health check-ups without having a long-term condition. In the Danish context, however, such cases are expected to be rare, as regular check-ups without medical indication are uncommon, and most regular appointments in primary or secondary care are part of structured follow-up for chronic conditions. We, therefore, consider the potential impact on our findings to be minimal. Second, potential misclassification of long-term conditions could influence the identification of latent classes, particularly if certain conditions are systematically under- or overreported. However, the indicator set spans 16 conditions across multiple domains, and the resulting classes were clinically plausible and consistent with prior literature. Conditions routinely managed in primary care – e.g., osteoarthritis, migraine/headache, tinnitus, cataract, mental health disorders, and musculoskeletal conditions – may be underreported in Danish administrative health data.⁵⁹ In such cases, survey-based data may capture a larger share of the affected population. In addition, applying calibrated survey weights reduces bias related to nonresponse and population composition, thereby improving the representativeness of both somatic and mental health indicators.^60,61 Future studies could employ broader disease measures, clinically validated data, and/or comparisons between register-based and survey-based disease classes. Third, although treatment burden was also measured using self-reported data prone to misclassification, we used a validated patient-reported outcome measure (MTBQ) specifically designed to capture the patient’s subjective experience. While inherently subjective, this is also its strength, as it reflects perceived effort rather than inferred workload. Fourth, our survey included only 16 long-term conditions, and the reported treatment burden may, therefore, reflect additional conditions not captured by the survey. The clinical management affecting the treatment burden of different conditions may vary by the type and severity of conditions and by involved healthcare providers across primary and secondary care, which we had no means to assess. Fifth, residual variations indicated by high bivariate residuals suggest potential unaccounted confounding effects, possibly causing differential item functioning or measurement non-invariance. Our stratified analyses by age provided some evidence supporting this hypothesis, though the direction of associations remained stable across subgroups. Further exploration of possible confounding effects in future studies is encouraged, including the potential role of interactions between social determinants, which was out of scope for this study. The study was conducted during the COVID-19 pandemic, possibly influencing treatment experiences, but similar findings from the 2017 survey reduce concerns about significant pandemic-related bias. Finally, although we applied a bias-adjusted three-step latent class approach that corrects for classification error, the model-dependent nature of class definitions remains a potential limitation. While our findings were robust to restrictions to individuals with high class assignment probability, future studies may explore the extent to which different class solutions affect substantive interpretations.

Conclusion

This study provides valuable guidance for clinicians and policymakers by identifying three complex multimorbidity groups at the population level characterised by diminished patient capacity and a markedly elevated risk of high treatment burden. Beyond confirming that greater disease counts are associated with higher burden, our findings indicate that the specific pattern of co-occurring conditions also matters. Incorporating multimorbidity profiles into patient risk assessments could inform healthcare planning, resource allocation, and the design of targeted interventions and integrated care pathways to reduce patient-perceived treatment burden.

Supplemental material

Supplemental material - Multimorbidity patterns and their associations with patient-perceived treatment burden: A latent class analysis of 14,344 Danish adults

Supplemental material for Multimorbidity patterns and their associations with patient-perceived treatment burden: A latent class analysis of 14,344 Danish adults by Marie Hauge Pedersen, Mathias Lasgaard, Camilla Palmhøj Nielsen, Anders Prior, Polly Duncan, Stine Schramm, Finn Breinholt Larsen in Journal of Multimorbidity and Comorbidity.

Footnotes

Acknowledgements

The Danish National Health Survey was collected by the five regions and the National Institute of Public Health, University of Southern Denmark. The MTBQ was developed by Professor Chris Salisbury and Dr Polly Duncan. Copyright (including the Danish version) belongs to the University of Bristol but is freely available for use under license. Please see for details. We thank them for their permission to use the MTBQ for this study.

Ethical considerations

Consent to participate

All invited survey participants were informed that participation was voluntary, that their survey data could be used for research purposes, including potential linkage with administrative register data, that reporting of study findings would be anonymised, ensuring that no individual reporting would be identifiable, and that full or partial survey completion constituted implied consent. The survey data were linked to administrative register data on age, sex, and country of origin using a unique personal identification number assigned to all Danish citizens, and all data were pseudonymised for analysis.

Consent for publication

Please see “Content to participate”.

Author Contributions

MHP had primary responsibility for all aspects of the study, including conceptualisation, data acquisition and curation, analysis, and drafting of the original manuscript. CPN, ML, FBL, AP, and PD supervised the process and provided valuable input. SS reviewed the study design and data curation plan before determining the final analytical approach. FBL and ML provided in-depth support for the analysis. All authors actively contributed to the revision of the original draft, participated in the interpretation of the results, and approved the final manuscript before submission.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the Institute of Public Health at Aarhus University; the Health Research Foundation of Central Denmark Region [grant number R86-A4212]; the Central Denmark Region Fund for Strengthening of Health Science; and the Central Denmark Region Health Survey. The Central Denmark Region Health Survey was conducted and funded by the Central Denmark Region.

Declaration of conflicting interests

The authors declared the following potential conflicts of interest with respect to the research, authorship, and/or publication of this article: PD developed and validated the original English version of the MTBQ. The authors declared no other potential conflicts of interest concerning this article’s research, authorship, and/or publication.

Data Availability Statement

Danish data protection rules do not allow us to share the individual-level survey and register data used for this study. Data access for research purposes may be obtained by applying the relevant data authorities (cf. Ethical considerations).*

Supplemental material

Supplemental Material for this article is available online.

ORCID iDs

Marie Hauge Pedersen

Mathias Lasgaard

Camilla Palmhøj Nielsen

Anders Prior

Stine Schramm

Finn Breinholt Larsen

Appendix

References

Salisbury

Johnson

Purdy

, et al. Epidemiology and impact of multimorbidity in primary care: a retrospective cohort study. Br J Gen Pract 2011; 61: e12-21. DOI: 10.3399/bjgp11X548929.

Jensen

HAR

Davidsen

Møller

, et al. Danskernes sundhed – Den nationale sundhedsprofil 2021 [The health of the Danes – The national health profile 2021]. København: Sundhedsstyrelsen; 2022.

Jensen

HAR

Davidsen

Ekholm

, et al. Danskernes sundhed – Den nationale sundhedsprofil 2017 [The health of the Danes – The national health profile 2017]. København: Sundhedsstyrelsen; 2018.

Frølich

Olesen

Kristensen

. Hvidbog om multisygdom. Dokumentation af multisygdom i det danske samfund - fra silotænkning til sammenhæng [White Paper on Multimorbidity: Documentation of Multimorbidity in Danish Society – From Silo Thinking to Integration]. 1st ed.. Denmark: Fjerritslev, 2017.

Barnett

Mercer

Norbury

, et al. Epidemiology of multimorbidity and implications for health care, research, and medical education: a cross-sectional study. Lancet 2012; 380: 37-43. DOI: 10.1016/S0140-6736(12)60240-2.

Mishra

Jones

. Evidence on multimorbidity from definition to intervention: An overview of systematic reviews. Ageing Res Rev 2017; 37: 53-68. DOI: 10.1016/j.arr.2017.05.003.

Jensen

Pedersen

Vestergaard

, et al. The impact of socioeconomic status and multimorbidity on mortality: a population-based cohort study. Clin Epidemiol 2017; 9: 279-289. DOI: 10.2147/clep.S129415.

Fortin

Soubhi

Hudon

, et al. Multimorbidity's many challenges. BMJ 2007; 334: 1016-1017. DOI: 10.1136/bmj.39201.463819.2C.

Larsen

Pedersen

Friis

, et al. A Latent Class Analysis of multimorbidity and the relationship to socio-demographic factors and health-related quality of life. A national population-based study of 162,283 Danish adults. PLoS One 2017; 12: e0169426. DOI: 10.1371/journal.pone.0169426.

10.

Prior

Vestergaard

Davydow

, et al. Perceived stress, multimorbidity, and risk for hospitalizations for ambulatory care-sensitive conditions: A population-based cohort study. Med Care 2017; 55: 131-139. DOI: 10.1097/mlr.0000000000000632.

11.

Wolff

Starfield

Anderson

. Prevalence, expenditures, and complications of multiple chronic conditions in the elderly. Arch Intern Med 2002; 162: 2269-2276. DOI: 10.1001/archinte.162.20.2269.

12.

Pedersen

Larsen

. Multisygdom i den danske befolkning – forekomsten af multisygdomsmønstre og sammenhængen med sociodemografiske faktorer og helbredsrelateret livskvalitet [Multimorbidity in the Danish population – the prevalence of multimorbidity patterns and the association with socio-demographic factors and health-related quality of life]. Aarhus: DEFACTUM, Region Midtjylland, 2022.

13.

May

Eton

Boehmer

, et al. Rethinking the patient: using Burden of Treatment Theory to understand the changing dynamics of illness. BMC Health Serv Res 2014; 14: 281. DOI: 10.1186/1472-6963-14-281.

14.

Vogeli

Shields

Lee

, et al. Multiple chronic conditions: prevalence, health consequences, and implications for quality, care management, and costs. J Gen Intern Med 2007; 22: 391-395.

15.

Bayliss

Steiner

Fernald

, et al. Descriptions of barriers to self-care by persons with comorbid chronic diseases. Ann Fam Med 2003; 1: 15-21.

16.

Schiøtz

Høst

Frølich

. Involving patients with multimorbidity in service planning: perspectives on continuity and care coordination. J Comorb 2016; 6: 95-102. DOI: 10.15256/joc.2016.6.81.

17.

Payne

Abel

Guthrie

, et al. The effect of physical multimorbidity, mental health conditions and socioeconomic deprivation on unplanned admissions to hospital: a retrospective cohort study. CMAJ 2013. DOI: 10.1503/cmaj.121349.

18.

Ørtenblad

Meillier

Jønsson

. Multi-morbidity: A patient perspective on navigating the health care system and everyday life. Chronic Illn 2018; 14: 271-282. DOI: 10.1177/1742395317731607.

19.

Eton

Ramalho de Oliveira

Egginton

, et al. Building a measurement framework of burden of treatment in complex patients with chronic conditions: a qualitative study. Patient Relat Outcome Meas 2012; 3: 39-49. DOI: 10.2147/PROM.S34681.

20.

Shippee

Shah

May

, et al. Cumulative complexity: a functional, patient-centered model of patient complexity can improve research and practice. J Clin Epidemiol 2012; 65: 1041-1051. DOI: 10.1016/j.jclinepi.2012.05.005.

21.

Smith

Wallace

Salisbury

, et al. A Core Outcome Set for Multimorbidity Research (COSmm). Ann Fam Med 2018; 16: 132-138. DOI: 10.1370/afm.2178.

22.

Farmer

Fenu

O'Flynn

, et al. Clinical assessment and management of multimorbidity: summary of NICE guidance. BMJ 2016; 354: i4843. DOI: 10.1136/bmj.i4843.

23.

Tran

V-T

Harrington

Montori

, et al. Adaptation and validation of the Treatment Burden Questionnaire (TBQ) in English using an internet platform. BMC Med 2014; 12: 109. DOI: 10.1186/1741-7015-12-109.

24.

Pedersen

Duncan

Lasgaard

, et al. Danish validation of the Multimorbidity Treatment Burden Questionnaire (MTBQ) and findings from a population health survey: a mixed-methods study. BMJ Open 2022; 12: e055276. DOI: 10.1136/bmjopen-2021-055276.

25.

Sav

Whitty

McMillan

, et al.

Treatment burden and chronic illness: Who is at most risk?

Patient 2016; 9: 559-569. DOI: 10.1007/s40271-016-0175-y.

26.

Gibbons

Kenning

Coventry

, et al. Development of a multimorbidity illness perceptions scale (MULTIPleS). PLoS One 2013; 8: e81852. DOI: 10.1371/journal.pone.0081852.

27.

Duncan

Murphy

Man

, et al. Development and validation of the Multimorbidity Treatment Burden Questionnaire (MTBQ). BMJ Open 2018; 8: e019413-012017-019413. DOI: 10.1136/bmjopen-2017-019413.

28.

Boyd

Wolff

Giovannetti

, et al. Healthcare task difficulty among older adults with multimorbidity. Med Care 2014; 52 Suppl 3: S118-125. DOI: 10.1097/MLR.0b013e3182a977da.

29.

Friis

Lasgaard

Pedersen

, et al. Health literacy, multimorbidity, and patient-perceived treatment burden in individuals with cardiovascular disease. A Danish population-based study. Patient Educ Couns 2019; 102: 1932-1938. DOI: 10.1016/j.pec.2019.05.013.

30.

Manning

Gagnon

. The complex patient: A concept clarification. Nurs Health Sci 2017; 19: 13-21. DOI: 10.1111/nhs.12320.

31.

Zullig

Whitson

Hastings

, et al. A Systematic Review of Conceptual Frameworks of Medical Complexity and New Model Development. J Gen Intern Med 2016; 31: 329-337. DOI: 10.1007/s11606-015-3512-2.

32.

Harrison

Britt

Miller

, et al. Examining different measures of multimorbidity, using a large prospective cross-sectional study in Australian general practice. BMJ Open 2014; 4: e004694. DOI: 10.1136/bmjopen-2013-004694.

33.

Dhafari

Pate

Azadbakht

, et al. A scoping review finds a growing trend in studies validating multimorbidity patterns and identifies five broad types of validation methods. J Clin Epidemiol 2024; 165: 111214. DOI: 10.1016/j.jclinepi.2023.11.004.

34.

Alvarez-Galvez

Ortega-Martin

Carretero-Bravo

, et al. Social determinants of multimorbidity patterns: A systematic review. Front Public Health 2023; 11: 1081518. DOI: 10.3389/fpubh.2023.1081518.

35.

Nichols

Taverner

Crowe

, et al. In simulated data and health records, latent class analysis was the optimum multimorbidity clustering algorithm. J Clin Epidemiol 2022; 152: 164-175. DOI: 10.1016/j.jclinepi.2022.10.011.

36.

Weller

Bowen

Faubert

. Latent Class Analysis: A guide to best practice. J Black Psychol 2020; 46: 287-311. DOI: 10.1177/0095798420930932.

37.

Christensen

Lau

Kristensen

, et al. The Danish National Health Survey: Study design, response rate and respondent characteristics in 2010, 2013 and 2017. Scand J Public Health 2022; 50: 180-188. DOI: 10.1177/1403494820966534.

38.

Pedersen

. The Danish civil registration system. Scand J Public Health 2011; 39: 22-25.

39.

Statistics Denmark . FOLK1A: Befolkningen den 1. i kvartalet efter område, køn, alder og civilstand [FOLK1A: Population on the 1st of the quarter by area, sex, age and marital status]. Available: https://www.statistikbanken.dk/FOLK1A [Accessed 10 February 2025].

40.

Friis

Jensen

Pedersen

, et al. Hvordan har du det? 2021 – Sundhedsprofil for region og kommuner [How are you? 2021 – Health profile for region and municipalities]. Aarhus: DEFACTUM, Region Midtjylland, 2022, Volume 1.

41.

Buratta

Frova

Gargiulo

, et al. Development of a common instrument for chronic physical conditions. In: Nosikov

Gudex

(eds) EUROHIS: Developing common instruments for health surveys. IOS Press, 2003, pp.21-34.

42.

Duncan

Scott

Dawson

, et al. Further development and validation of the Multimorbidity Treatment Burden Questionnaire (MTBQ). BMJ Open 2024; 14: e080096. DOI: 10.1136/bmjopen-2023-080096.

43.

OECD, Eurostat and UNESCO Institute for Statistics . ISCED 2011 operational manual: Guidelines for classifying national education programmes and related qualifications. Paris: OECD Publishing, 2015.

44.

Sherbourne

Stewart

. The MOS social support survey. Soc Sci Med 1991; 32: 705-714. DOI: 10.1016/0277-9536(91)90150-b.

45.

Maindal

Kayser

Norgaard

, et al. Cultural adaptation and validation of the Health Literacy Questionnaire (HLQ): robust nine-dimension Danish language confirmatory factor model. SpringerPlus 2016; 5: 1232-1232. DOI: 10.1186/s40064-016-2887-9.

46.

Osborne

Batterham

Elsworth

, et al. The grounded psychometric development and initial validation of the Health Literacy Questionnaire (HLQ). BMC Public Health 2013; 13: 658.

47.

Särndal

C-E

Lundström

. Estimation in surveys with nonresponse. West Sussex, England: Wiley, 2005.

48.

Nylund-Gibson

Grimm

Masyn

. Prediction from latent classes: A demonstration of different approaches to include distal outcomes in mixture models. Struct Equ Modeling 2019; 26: 967-985. DOI: 10.1080/10705511.2019.1590146.

49.

Bakk

Kuha

. Relating latent class membership to external variables: An overview. Br J Math Stat Psychol 2021; 74: 340-362. DOI: 10.1111/bmsp.12227.

50.

Nylund

Asparouhov

Muthén

. Deciding on the number of classes in latent class analysis and growth mixture modeling: A Monte Carlo simulation study. Struct Equ Modeling 2007; 14: 535-569. DOI: 10.1080/10705510701575396.

51.

Masyn

. Latent class analysis and finite mixture modeling. In: Little

(ed) The Oxford handbook of quantitative methods: Statistical analysis, Vol 2. New York, NY, US: Oxford University Press; 2013, pp.551-611.

52.

Vermunt

. Latent class modeling with covariates: Two improved three-step approaches. Polit Anal 2010; 18: 450-469. DOI: 10.1093/pan/mpq025.

53.

Bolck

Croon

Hagenaars

. Estimating latent structure models with categorical variables: One-step versus three-step estimators. Polit Anal 2004; 12: 3-27. DOI: 10.1093/pan/mph001.

54.

Bakk

Tekle

Vermunt

. Estimating the association between latent class membership and external variables using bias-adjusted three-step approaches. Sociol Methodol 2013; 43: 272-311. DOI: 10.1177/0081175012470644.

55.

Bakk

Oberski

Vermunt

. Relating latent class assignments to external variables: Standard errors for correct inference. Polit Anal 2014; 22: 520-540. DOI: 10.1093/pan/mpu003.

56.

Vermunt

Magidson

. Technical Guide for Latent GOLD 5.1: Basic, Advanced, and Syntax. Belmont, MA: Statistical Innovations Inc; 2016.

57.

Larsen

Pedersen

Lasgaard

, et al. Hvordan har du det? 2017 – Sundhedsprofil for region og kommuner [How are you? 2017 – Health profile for region and municipalities] (Volume 1). Aarhus: DEFACTUM, Region Midtjylland; 2018.

58.

Schramm

Møller

Tolstrup

, et al. Effects of individual and parental educational levels on multimorbidity classes: a register-based longitudinal study in a Danish population. BMJ Open 2022; 12: e053274. DOI: 10.1136/bmjopen-2021-053274.

59.

Hvidberg

Johnsen

Davidsen

, et al. A Nationwide Study of Prevalence Rates and Characteristics of 199 Chronic Conditions in Denmark. Pharmacoecon Open 2020; 4: 361-380. DOI: 10.1007/s41669-019-0167-7.

60.

Momen

Lasgaard

Weye

, et al. Representativeness of survey participants in relation to mental disorders: a linkage between national registers and a population-representative survey. Int J Popul Data Sci 2022; 7. DOI: 10.23889/ijpds.v7i4.1759.

61.

Jensen

HAR

Lau

Davidsen

, et al. The impact of non-response weighting in health surveys for estimates on primary health care utilization. Eur J Public Health 2022; 32: 450-455. DOI: 10.1093/eurpub/ckac032.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

1.12 MB