Identifying subgroups of high-need, high-cost, chronically ill patients in primary care: A latent class analysis.

INTRODUCTION: Segmentation of the high-need, high-cost (HNHC) population is required for reorganizing care to accommodate person-centered, integrated care delivery. Therefore, we aimed to identify and characterize relevant subgroups of the HNHC population in primary care by using demographic, biomedical, and socioeconomic patient characteristics.
METHODS: This was a retrospective cohort study within a Dutch primary care group, with a follow-up period from September 1, 2014 to August 31, 2017. Chronically ill patients were included in the HNHC population if they belonged to the top 10% of care utilizers and/or suffered from multimorbidity and had an above-average care utilization. In a latent class analysis, forty-one patient characteristics were initially used as potential indicators of heterogeneity in HNHC patients' needs.
RESULTS: Patient data from 12 602 HNHC patients was used. A 4-class model was considered statistically and clinically superior. The classes were named according to the characteristics that were most dominantly present and distinctive between the classes (i.e. mainly age, household position, and source of income). Class 1 ('older adults living with partner') included 39.3% of patients, class 2 ('older adults living alone') included 25.5% of patients, class 3 ('middle-aged, employed adults with family') included 23.3% of patients, and class 4 ('middle-aged adults with social welfare dependency') included 11.9% of patients. Diabetes was the most common condition in all classes; the second most prevalent condition differed between osteoarthritis in class 1 (21.7%) and 2 (23.8%), asthma in class 3 (25.3%), and mood disorders in class 4 (23.1%). Furthermore, while general practitioner (GP) care utilization increased during the follow-up period in the classes of older adults, it remained relatively stable in the middle-aged classes.
CONCLUSIONS: Although the HNHC population is heterogeneous, distinct subgroups with relatively homogeneous patterns of mainly demographic and socioeconomic characteristics can be identified. This calls for tailoring care and increased attention for social determinants of health.

Introduction

Due to increasing numbers of chronically ill patients, in particular with multimorbidity, and rising health care costs, Western health care systems are faced with challenges to deliver high-quality, person-centered, and sustainable care [1-3]. In response to these developments, accountable care organizations (ACOs) were introduced in the United States several years ago [4-6]. Within ACOs, a value-based payment system is designed to incentivize providers to share accountability for the quality and cost of care for a defined population [6-8]. Likewise, more than a decade ago, ‘care groups’ were first introduced in Dutch primary care. In line with the ACOs, care groups unite providers, mostly general practitioners (GPs), with shared responsibility for all assigned patients receiving care for a specific chronic condition from a value-based bundled payment approach [5, 9]. These initiatives show that, similar to the US, the Netherlands aims to achieve more value-based care.

If health systems aim to increase the value of delivered care, it is crucial to focus on the population with the highest care use as they offer the largest potential for achieving improved value [10, 11]. This population with a disproportionately high care use is also referred to as the high-need, high-cost (HNHC) population [10, 12]. The identification of the HNHC population, as a subgroup of the total population, is embedded in the approach of population segmentation, which is defined as the division of a specific population into homogeneous subgroups with distinct needs and (health) characteristics [13-15]. A closely related concept in which principles of segmentation are applied, pertains to the concept of ‘population (health) management’ (PM),[16] as a way to promote ‘population health’ [17, 18]. Within population health, the focus is on the health outcomes of subgroups rather than individuals, by taking into account a large variety of determinants of health (i.e. physical, mental, social) [17, 18]. PM strategies generally aim to improve health needs of defined subgroups along ‘the continuum of health and well-being’, and aim to integrate services across multiple domains [16]. As such, PM strategies can be used to tailor interventions to the care needs of specific subgroups of patients, which is assumed to lead towards improving individual patients’, and providers’ experiences as well as population outcomes and cost (Quadruple Aim[19]).

With the growing availability of digital patient data, studies have identified common biomedical characteristics of the HNHC population, such as the high prevalence of (co-occurring) chronic conditions and mental illness [20, 21]. At the same time, studies have suggested that the HNHC population is diverse, not only in terms of patients’ biomedical but also in their demographic and socioeconomic profiles [10, 20, 21]. These findings underline the importance of social determinants of health within the HNHC population. Yet, population segmentation studies have predominantly focused on specific populations, such as older adults [22-24] and Medicaid beneficiaries [25], and mainly characterized the identified patient subgroups by their biomedical characteristics (i.e., chronic diagnoses) [22-27]. Therefore, the main aim of this study was to identify and characterize, by means of latent class analysis (LCA), clinically relevant subgroups of the HNHC population in primary care, defined by demographic, biomedical, and socioeconomic patient characteristics as well as care utilization.

Materials and methods

Setting

This retrospective cohort study was conducted at a (primary) care group in the northern region of the Netherlands, covering 130 general practices. This care group was founded in 2009 and currently has bundled payment contracts with health insurers for the delivery of several disease management programs, including for patients with type 2 diabetes mellitus, COPD, and cardiovascular risks.

As this study used retrospective data and did not intervene into people’s life or impose rules, no formal ethical approval was required (project number 164111), in line with the Dutch Medical Research (Human Subjects) Act.

Data sources

All general practices connected to the care group were invited to extract and provide individual-level patient data from their electronic health records (EHRs). The EHR data covered 4.5 years: baseline was on September 1, 2014; the follow-up period covered three years (from September 1, 2014 to August 31, 2017). Furthermore, the EHR data were linked on the individual patient level to socioeconomic data (e.g., source of income) and health care claims data (e.g., pharmaceutical costs). Socioeconomic data were retrieved from Statistics Netherlands, which is involved in the collection, preparation, and publication of statistics on behalf of the Dutch government, science and commercial sector [28]. Claims data were retrieved from the health care information center ‘Vektis’, which collects and manages all claims under the Dutch Healthcare Insurance Act [29]. To ensure data confidentiality and safety, a third trusted party was involved in the provision of a pseudonymized version of the data set to the researchers.

Participants

We selected a cohort of chronically ill patients, limited to those with a full EHR registration over the 4.5-year research period. Patients were considered chronically ill if they had registered at least one GP consultation in the 1.5 years before baseline related to one of 28 conditions defined as chronic (see Table 1) [30, 31]. Chronically ill patients were included in the HNHC population if they belonged to the top 10% of care utilizers (over follow-up period) and/or suffered from multimorbidity and had an above-average care utilization (over follow-up period). The first criterion was applied as this is one of the commonly used thresholds for identifying HNHC patients according to previous studies [20, 32, 33]. The second criterion was applied because multimorbidity brings along a challenging complexity to the organization of care, especially in light of the current single-disease management programs for single chronic conditions [2, 3]. Furthermore, care utilization was measured as the total number of GP consultations weighted by the required time investment per type of consultation (i.e. 0.5 for telephone or e-mail consultation, 1.0 for regular consultation, 2.0 for extended regular consultation, 1.5 for home visit, 2.5 for extended home visit), determined by the Netherlands Institute for Health Services Research [34]. As the weighting factors based on time investment are related to costs [35], the patients selected for this study can be considered high-need, high-cost in primary care.

Table 1

Baseline characteristics of the HNHC population (n = 12 602).

Patient characteristics
	n (%)	Missing, n (%)
Demographic characteristics
Sex		0
Male	4495 (35.67)
Female	8107 (64.33)
Age, mean (SD)a	67.55 (14.80)	0
Household position		0
Child living at home	141 (1.12)
Single adult	3773 (29.94)
Partner with children at home	1515 (12.02)
Partner without children at home	6245 (49.56)
Single parent	403 (3.20)
Member of collective household	371 (2.94)
Other	154 (1.22)
Age of children living at parental home		0
≤12	388 (3.08)
>12	1752 (13.90)
No children living at home	10 462 (83.02)
Biomedical characteristics
Type of chronic condition(s)		0
Only physical	10 060 (79.83)
Only mental	436 (3.46)
Combination of both	2106 (16.71)
Number of chronic conditions, mean (SD)	2.23 (0.93)	0
Prevalence of 28 chronic conditions		0
Chronic alcohol abuse	163 (1.29)
Endocardial conditions, valvular conditions	298 (2.36)
Congenital cardiovascular anomaly	25 (0.20)
HIV/AIDS	9 (0.07)
Anxiety disorders	649 (5.15)
Asthma	2142 (17.00)
Stroke (including TIA)	986 (7.82)
Chronic obstructive pulmonary disease (COPD)	2218 (17.60)
Chronic back or neck disorder	2033 (16.13)
Coronary heart diseases	1725 (13.69)
Dementia including Alzheimer’s	172 (1.36)
Diabetes mellitus	4925 (39.08)
Epilepsy	181 (1.44)
Hearing disorders	679 (5.39)
Visual disorders	1694 (13.44)
Heart failure	659 (5.23)
Heart arrhythmia	1446 (11.47)
Cancer	2032 (16.12)
Migraine	395 (3.13)
Osteoporosis	737 (5.85)
Burnout	452 (3.59)
Osteoarthritis	2360 (18.73)
Personality disorders	120 (0.95)
Rheumatoid arthritis	433 (3.44)
Schizophrenia	53 (0.42)
Mood disorders	1380 (10.95)
Mental retardation	48 (0.38)
Parkinson’s disease	136 (1.08)
Socioeconomic characteristics
Housing situation		12 (0.10)
Owner-occupied	6777 (53.78)
Rentedb	5813 (46.13)
Source of income		0
Paid workc	1974 (15.66)
Social welfare or unemployment benefits	1838 (14.58)
Pension benefits	8156 (64.72)
Without incomed	634 (5.03)
Number of people in a household with an individual income		26 (0.21)
1	4594 (36.45)
>1	7982 (63.34)
Household dependence on social security payments, mean (SD)	11.63 (25.44)	346 (2.75)
Paid interest over debts, mean (SD)	48.89 (782.63)	20 (0.16)
Care utilization
Pharmaceutical costs		16 (0.13)
≤€500	4773 (37.87)
>€500 and ≤€1500	5122 (40.64)
>€1500	2691 (21.35)
GP care utilization before baseline, mean (SD)	29.97 (18.50)	0

a For continuous variables, mean (SD) is reported.

b Includes members of collective households.

c Includes employees, entrepreneurs, and managers.

d Includes students with and without individual income.

Variables

Forty-one patient characteristics were initially used as potential indicators of heterogeneity in HNHC patients’ needs in the LCA. These characteristics were included based on scientific studies describing these characteristics as relevant in relation to (high) care utilization [12, 36]. Demographic characteristics were measured at baseline and included patients’ sex, age (in years), household position (child living at home, single adult, partner with children at home, partner without children at home, single parent, member of a collective household, other), and age of children living at parental home (≤12 years, >12 years (i.e. the age that they generally leave elementary school), no children living at home). Biomedical characteristics were also measured at baseline and included patients’ chronic disease diagnoses based on GP care use related to the chronic disease in the 1.5 years before baseline, type of chronic condition(s) (only physical, only mental, or a combination of both), and number of chronic conditions (1 to 28). All socioeconomic characteristics, except for source of income, were measured over the year 2014 and included patients’ (household) housing situation (owner-occupied, rented), number of people in a household with an individual income (1, >1), household dependence on social security payments as proportion of gross household income (0% to 100%), and paid interest over debts (in euros, excluding mortgage or debts related to renovating personal property). Source of income (paid work, social welfare or unemployment benefits, pension benefits, without income) was measured at baseline. Care utilization characteristics included GP care utilization on baseline (number of registred GP consultations) and patients’ pharmaceutical costs (≤€500, >€500 and ≤€1500, >€1500) which were measured over 2014.

Data analysis

Data were validated and checked for outliers and missing values. We employed LCA, which is a sophisticated analysis technique to capture heterogeneity in the HNHC population’s needs by the smallest number of unobserved homogeneous classes [37]. Furthermore, LCA is a person-oriented analysis technique [37] which aims to identify classes of individuals with similar patterns of, in the current study, (correlated) personal factors relevant to health care utilization. Initially, the LCA was run using all 41 patient characteristics (see Table 1) in order to explore the potential to identify clinically relevant subgroups. Furthermore, the analysis was conducted with a maximum likelihood estimator with robust standard errors (MLR). Missing values were handled by the default option in the Mplus software (version 8.1). To test whether the missing values were completely at random (MCAR), a MCAR Pearson-Chi Square and Likelihood Ratio Chi-Square test (P < .05) was computed. Additionally, the number of random starts values was increased several times to prevent problems related to nonconvergence or local maxima.[38] By stepwise increasing the number of classes, starting with a 1-class model, and comparing various statistical indicators and clinical relevance, we decided on the final model. Statistical indicators for model fit included the Akaike Information Criterion (AIC), [39, 40] Bayesian Information Criterion (BIC), [41] bootstrapped likelihood ratio test (BLRT), [42] and entropy score. Lower values on AIC and BIC indicated better model fit; significant p-values on the BLRT showed dominance of the k class model, compared to the k-1 class model. The entropy score gave an indication of classification certainty, using a cutoff score of at least 0.8, indicating high classification certainty [38]. The BIC and BLRT were considered most important in deciding on the best model as these outperform other statistical indicators [43].

Besides statistical indicators, clinical relevance of the model was a key factor, as the model should support daily clinical practice [15]. Also, the size of the classes within the model was taken into account (also reffered to as substantiality) [15]. A model with classes including at least 10% of HNHC population was considered substantial to counterbalance efforts to tailor interventions in daily practice. Although we aimed to maintain the largest variety of patient characteristics, the model was made more parsimonious after identifying a clinically relevant model. Thus, we removed any variables that did not contribute to the division in clinically relevant classes, significantly deteriorated the model fit, and/or were regarded as being of less added value based on internal clinical insight. Patients in each class of the final model were described in terms of the probability of having a given patient characteristic. In line with previous studies using LCA, probabilities of 70% to 100% were considered high, probabilities of 40% to 69% moderate, and probabilities of less than 40% low [44, 45]. The continuous variables were described by their estimated mean (SE). Furthermore, each class was described in their top five of chronic conditions at baseline and mean GP care utilization (i.e. mean number of weighted GP consultations) over the follow-up period.

Results

Baseline characteristics

A total of 63 general practices (48.5%) participated. The complete data set included individual-level data from 58 551 chronically ill patients, of whom 12 602 patients (21.5%) met the inclusion criteria for the study (i.e., were considered HNHC). Baseline characteristics of the HNHC population, including number (%) of missing values per characteristic, are shown in Table 1. Patients’ mean (SD) GP care utilization over the follow-up period was 66.9 contacts (33.3).

Latent class analysis

A 4-class model was considered statistically and clinically superior. The 4-class model had a low value on BIC, a significant BLRT (P < .001), high entropy score (0.973), and each class was sufficiently substantial by including at least 10% of the HNHC population (see Table 2). Although the 5-class model was statistically superior to the 4-class model, it included two classes with less than 10% of the HNHC population and resulted in less relevant and distinct classes compared to the 4-class model. More specifically, a 5-class model largely maintained three of the four classes of the 4-class model and subdivided the fourth and smallest class of the 4-class model into two smaller classes which were relatively indistinct from each other.

Table 2

Statistical indicators and relative class sizes for models with increasing numbers of latent classes.

	1-class model	2-class model	3-class model	4-class model	5-class model
Loglikelihood	-183,726.630	-172,407.886	-164,350.740	-159,286.403	-154,427.535
AICa	367,493.259	344,893.772	328,817.480	318,726.806	309,047.071
BICb	367,642.092	345,183.995	329,249.094	319,299.810	309,761.466
Entropy	n/a	0.981	0.974	0.973	0.977
BLRTc	n/a	P < .001	P < .001	P < .001	P < .001
Relative class size	n/a	86.62/13.38	64.51/23.58/11.90	39.30/25.51/23.31/11.87	38.18/25.31/18.48/9.21/8.82

aAIC refers to Akaike Information Criterion

bBIC refers to Bayesian Information Criterion

cBLRT refers to bootstrapped likelihood ratio test

Table 3 shows the final model, which includes nine of the initially used 41 patient characteristics and the probabilities of having each patient characteristic, given class membership (see also Fig 1). This means that the following variables were excluded in the final LCA due to less statistical relevance: age of children living at parental home, number of chronic conditions, prevalence of 28 chronic conditions, paid interest over debts, GP care utilization on baseline.The MCAR Pearson-Chi Square and Likelihood Ratio Chi-Square test showed that values were missing completely at random (P <0.001). As the entropy score was high, we report the final class counts and proportions for the latent classes that are based on their most likely latent class membership. Class 1 (n = 4953; 39.3%) had a mean (SE) age of 74.5 years (0.10), had a high probability (0.91) of having a partner but no children at home, and a high probability (0.98) of receiving pension benefits. Based on these dominant characteristics, class 1 was named ‘older adults living with partner’. Class 2 (n = 3215; 25.5%) had a mean (SE) age of 78.8 years (0.15), had a high probability (0.92) of being single, and a high probability (0.99) of receiving pension benefits. Based on these dominant characteristics, class 2 was named ‘older adults living alone’. Class 3 (n = 2938; 23.3%) had a mean (SE) age of 51.0 years (0.24) and had a high probability of having a partner with or without children at home (0.82). In terms of socioeconomic status, members of class 3 had a moderate probability (0.62) of having paid work. Based on these dominant characteristics, class 3 was named ‘middle-aged, employed adults with family’. Class 4 (n = 1496; 11.9%) had a mean (SE) age of 52.2 years (0.32). With regard to household position, members of class 4 had a low probability (0.34) of being single and a low probability (0.33) of having a partner but no children at home. In terms of socioeconomic status, members of class 4 had a high probability (0.84) of receiving social welfare or unemployment benefits. Based on these dominant characteristics, class 4 was named ‘middle-aged adults with social welfare dependency’. See also S1 File for a description of typical qualitative personas who characterize the four classes.

Table 3

Probabilities of having the (categorical) patient characteristic, given class membership, for each class within the final 4-class model.

Patient characteristics	Probability (SE)
	Class 1	Class 2	Class 3	Class 4
	(n = 4953)	(n = 3215)	(n = 2938)	(n = 1496)
Demographic characteristics
Sex
Male	0.481 (0.01)	0.202 (0.01)	0.290 (0.01)	0.404 (0.01)
Female	0.519 (0.01)	0.798 (0.01)	0.710 (0.01)	0.596 (0.01)
Age, mean (SE)a	74.47 (0.10)	78.78 (0.15)	51.01 (0.24)	52.22 (0.32)
Household position
Child living at home	0.001 (0.00)	0.000 (0.00)	0.034 (0.00)	0.027 (0.00)
Single adult	0.010 (0.00)	0.917 (0.08)	0.092 (0.01)	0.341 (0.01)
Partner with children at home	0.034 (0.00)	0.000 (0.00)	0.391 (0.01)	0.144 (0.01)
Partner without children at home	0.905 (0.01)	0.000 (0.00)	0.424 (0.01)	0.332 (0.01)
Single parent	0.022 (0.00)	0.002 (0.00)	0.044 (0.00)	0.105 (0.01)
Member of a collective household	0.010 (0.00)	0.080 (0.07)	0.002 (0.00)	0.039 (0.01)
Other	0.019 (0.00)	0.001 (0.00)	0.013 (0.00)	0.012 (0.00)
Biomedical characteristics
Type of chronic condition
Only physical	0.891 (0.00)	0.863 (0.01)	0.664 (0.01)	0.610 (0.01)
Only mental	0.008 (0.00)	0.014 (0.00)	0.077 (0.01)	0.085 (0.01)
Combination of both	0.101 (0.00)	0.123 (0.01)	0.259 (0.01)	0.306 (0.01)
Socioeconomic characteristics
Housing situation
Owner-occupied	0.637 (0.01)	0.343 (0.01)	0.723 (0.01)	0.272 (0.01)
Rented	0.363 (0.01)	0.657 (0.01)	0.277 (0.01)	0.728 (0.01)
Source of income
Paid work	0.018 (0.00)	0.007 (0.00)	0.621 (0.01)	0.047 (0.01)
Social welfare or unemployment benefits	0.002 (0.00)	0.002 (0.00)	0.195 (0.01)	0.844 (0.01)
Pension benefits	0.981 (0.00)	0.990 (0.00)	0.000 (0.00)	0.042 (0.01)
Without income	0.000 (0.00)	0.001 (0.00)	0.184 (0.01)	0.068 (0.01)
Number of people with an individual income in a household
1	0.026 (0.00)	0.969 (0.01)	0.218 (0.01)	0.483 (0.01)
>1	0.974 (0.00)	0.031 (0.01)	0.782 (0.01)	0.517 (0.01)
Household dependence
on social security payments, mean (SE)a	1.28 (0.09)	0.35 (0.06)	9.28 (0.33)	75.81 (0.64)
Care utilization
Pharmaceutical costs
≤€500	0.353 (0.01)	0.318 (0.01)	0.513 (0.01)	0.340 (0.01)
>€500 and ≤€1500	0.439 (0.01)	0.423 (0.01)	0.349 (0.01)	0.378 (0.01)
>€1500	0.208 (0.01)	0.259 (0.01)	0.138 (0.01)	0.282 (0.01)

a For continuous variables, mean (SE) is reported

Fig 1

Probabilities of having the (categorical) patient characteristic, given class membership, for each class within the final 4-class model.

In terms of the top five chronic conditions per class at baseline (see Fig 2), diabetes mellitus was most common in each of the four classes, with prevalence ranging from 30.5% in class 3 to 43.4% in class 1. The second most prevalent condition differed between osteoarthritis in class 1 (21.7%) and 2 (23.8%), asthma in class 3 (25.3%), and mood disorders in class 4 (23.1%).

Fig 2

Top five of chronic conditions (%) per class within the final 4-class model.

With regard to GP care utilization of the classes over the follow-up period (see Fig 3), class 2 showed the highest mean care utilization. Both classes with the older adults showed the largest mean (SD) increase in care utilization over time—from 9.8 (6.9) in the first to 11.7 (8.7) in the sixth half year and from 11.5 (8.3) in the first to 14.0 (10.5) in the sixth half year—while the classes with the middle-aged adults were more stable over time—from 10.1 (7.1) in the first to 10.7 (8.2) in the sixth half year and from 11.3 (8.0) in the first to 12.1 (9.5) in the sixth half year.

Fig 3

GP care utilization measured over the follow-up period for each class within the final 4-class model.

Discussion

The present study suggests that the HNHC population in primary care is a heterogeneous population, which can be divided into four subgroups with distinct patterns of particularly demographic and socioeconomic characteristics. Main differences between the subgroups were found in demographic and socioeconomic factors (i.e. age, household position, and source of income). In terms of chronic conditions, the subgroups with older adults most frequently suffered from physical and age-related conditions (e.g. osteoarthritis, cancer), while the middle-aged subgroups most frequently had conditions more typically found in relatively younger people (i.e., asthma and mood disorders). Furthermore, while the subgroups with older adults showed an increase in mean care utilization over time, the middle-aged subgroups showed a more stable pattern over time. In addition, class 2 (‘older adults living alone’) showed the highest mean care utilization over time. This finding corresponds with a study of Dreyer, Steventon [36] who showed that living alone is associated with higher care utilization in older adults.

The current study indicates that the sex distribution within the HNHC population, as well as in three of the four identified subgroups, is unbalanced: more than 64% of the HNHC population is female. In the current person-oriented analysis, unlike in a variable-oriented analysis, there is no assessment of relations between variables including corrections for confounders. Rather, the current analysis has focused on identifying subgroups based on patterns of variables within individual patients. One possible explanation for the unbalanced population in terms of sex is that women typically get older and, as a result, are overrepresented among the older aged HNHC patients compared to men. In addition, scientific studies have found that women have significantly higher consultation rates compared to men, but particularly during working years [46, 47].

Our findings show that the HNHC population is a demographically and socioeconomically diverse population and includes not only older adults but also many middle-aged people. To date, studies have predominantly focused on (biomedical) segmentation in populations of older adults: an example is the recent Embrace study,[48] which identified three risk profiles for older adults. In line with the demographic heterogeneity found in our study, a study by Wammes, Tanke [12] found that many high-cost patients (in the Dutch curative health system) are not older than 65 years of age. Supporting our approach, the authors [12] emphasized the need for studying the general population with extensive data and targeting interventions toward high-cost patients of various ages. Furthermore, our findings suggest that middle-aged HNHC patients are generally characterized by more socioeconomic vulnerability (e.g., dependence on social welfare) and a higher prevalence of mental conditions (e.g., mood disorders) than are older HNHC patients. These findings add to an increasing awareness about the importance of social and context-related determinants of health [25, 49, 50]. First, Shadmi [51] suggests broadening the understanding and measurement of multimorbidity by including a large variety of health and health-related aspects (e.g., social, cultural, and economic background of populations) that correlate with multimorbidity. In addition, corresponding to our finding that current segmentation often lacks inclusion of relevant demographic and socioeconomic characteristics, the study by Chin-Yee, Subramanian [52] and Khoury, Iademarco [53] also argued that adding environmental and social characteristics (a rather “population perspective”) to the genetic profiling in precision medicine can be of added value to public health.

With the growing recognition of the effectiveness of segmentation for patient-centered interventions, [54] the segmentation conducted in the present study can guide clinical practice toward more integrated and person-centered care. By gathering insight into demographic characteristics other than age and gender (e.g., household position) as well as the socioeconomic context of patients (e.g., main source of income), clinical practice in primary care can be attuned to a more holistic view of patients. This view can suggest potentially relevant goals, interventions, and professionals (within primary care and in cooperation with other disciplines), which can be further discussed in a shared decision-making process with the patient. Such an approach can be inspired by the ‘Bridges to Health’ model, [55] which aims to systematically connect priority concerns, major components of health care, and goals for health care within identified population segments [56]. Thus, while older adults living alone might benefit from increased social support, middle-aged adults with social welfare dependency might rather benefit from financial and mental support. As such, this segmentation approach can serve as a starting point for more biopsychosocial attention and can inform the discussion of tailored interventions with the patient [56]. However, the individual consultation is still key to assess personal needs and preferences with a patient during a consultation, and agree on an individual treatment course.

Further research, in particular qualitative inquiry, is necessary to identify the most important concerns and components of health care per HNHC subgroup. In addition, the current study has focused on HNHC patients in primary care, which is widely considered the most suitable medical home for chronically ill patients [57]. Although as a result, our findings are mainly useful for improvement of primary care management, there is some evidence that patients with a disproportionately high use of primary care resources also account for significantly high(er) costs in specialist care [58, 59]. For policy making, the subgroups can also help to give insight into the distribution of the patient population over the identified subgroups within certain geographical areas and help to efficiently target resources. In more urban areas, for example, the middle-aged subgroups might be larger than in rural areas.

One of the most important strengths of the current study is the relatively large set of individual-level patient data, with a variety of patient characteristics. A second strength is the use of the model-based analysis technique LCA, which offers a large set of statistical indicators to decide on the best-fitting model and ways to cope with issues of local maxima and nonconvergence [38]. The study also has some limitations. First, individual level data of the non-participating practices were not available in this study. This hampered a direct comparison of participating practices (n = 63; 48.5% of care group) with non-participating practices in order to assess representativeness of the sample. However, particular patient characteristics (i.e. sex, age, household position, and source of income) of the sample were compared to the patient characteristics of the general population in the northern region of the Netherlands that is covered by the primary care group. This comparison showed that the sample is largely similar in patient characteristics to the general population. For example, 50.8% of the sample is female; 50.5% of the general population is female, 20.1% of the sample receives pension benefits; 22.1% of the general population receives pension benefits. Second, EHRs typically include incomplete registrations and may have limited data quality. Nevertheless, the quality of registrations was checked and validated, and the (categorical) missing values were found to be MCAR. Third, the data set included patients who can be considered dependent, as they belonged to the same household. A sensitivity analysis with only completely independent observations showed the same division among classes, implying a negligible effect of the dependent observations on the identification of subgroups. Fourth, only patients with a full EHR registration over the research period were included. This has excluded specific types of patients, such as patients who died before the end of the follow-up period. It is possible that the excluded patients would have been identified as a separate ‘near end of life’ HNHC subgroup, as identified by some previous population segmentation studies as well [24, 55, 60]. Nevertheless, specific payments arrangements are already in place in Dutch primary care for this patient population who is near the end of life and needs (expensive) palliative care. Fifth, generalizability of the subgroups may be limited, as the data set was retrieved from a specific Dutch region with limited ethnic/cultural diversity and a relatively aged population, compared to the Dutch average. In future research, the generalizability of the subgroups needs to be determined.

Conclusions

Despite the heterogeneity of the HNHC population, distinct subgroups with relatively homogeneous patterns of particularly demographic and socioeconomic characteristics can be identified. This study adds to the increasing awareness of the demographic and socioeconomic heterogeneity of the HNHC population, in addition to biomedical diversity. To accommodate person-centered, integrated care delivery, the identified classes need to be connected to tailored care (i.e. concerns, components, goals). This connection can be inspired by the proposed strategies within the ‘Bridges to Health’ model [55].

Description of typical qualitative personas who characterize the four identified classes.

(DOCX)

Click here for additional data file.

24 Oct 2019

PONE-D-19-25812

Identifying subgroups of high-need, high-cost, chronically ill patients in primary care: A latent class analysis

PLOS ONE

Dear Miss Smeets,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

The manuscript has been assessed by two reviewers. Their comments are appended below. The reviewers have raised significant concerns about the manuscript.

We would appreciate receiving your revised manuscript by Dec 08 2019 11:59PM. When you are ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter.

To enhance the reproducibility of your results, we recommend that if applicable you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. For instructions see: http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). This letter should be uploaded as separate file and labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. This file should be uploaded as separate file and labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. This file should be uploaded as separate file and labeled 'Manuscript'.

Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out.

We look forward to receiving your revised manuscript.

Kind regards,

Bruno Pereira Nunes, Ph.D.

Academic Editor

PLOS ONE

Journal Requirements:

When submitting your revision, we need you to address these additional requirements.

1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at http://www.journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and http://www.journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

2. We note that you have indicated that data from this study are available upon request. PLOS only allows data to be available upon request if there are legal or ethical restrictions on sharing data publicly. For information on unacceptable data access restrictions, please see http://journals.plos.org/plosone/s/data-availability#loc-unacceptable-data-access-restrictions.

In your revised cover letter, please address the following prompts:

a) If there are ethical or legal restrictions on sharing a de-identified data set, please explain them in detail (e.g., data contain potentially identifying or sensitive patient information) and who has imposed them (e.g., an ethics committee). Please also provide contact information for a data access committee, ethics committee, or other institutional body to which data requests may be sent.

b) If there are no restrictions, please upload the minimal anonymized data set necessary to replicate your study findings as either Supporting Information files or to a stable, public repository and provide us with the relevant URLs, DOIs, or accession numbers. Please see http://www.bmj.com/content/340/bmj.c181.long for guidelines on how to de-identify and prepare clinical data for publication. For a list of acceptable repositories, please see http://journals.plos.org/plosone/s/data-availability#loc-recommended-repositories.

We will update your Data Availability statement on your behalf to reflect the information you provide.

3. Thank you for stating the following in the Financial Disclosure section:

'The current study was funded by primary care group 'HZD', the Netherlands (https://www.hzd.nu). The funding agency had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.'

We note that you received funding from a commercial source: HZD.

Please provide an amended Competing Interests Statement that explicitly states this commercial funder, along with any other relevant declarations relating to employment, consultancy, patents, products in development, marketed products, etc.

Within this Competing Interests Statement, please confirm that this does not alter your adherence to all PLOS ONE policies on sharing data and materials by including the following statement: "This does not alter our adherence to PLOS ONE policies on sharing data and materials.” (as detailed online in our guide for authors http://journals.plos.org/plosone/s/competing-interests).  If there are restrictions on sharing of data and/or materials, please state these. Please note that we cannot proceed with consideration of your article until this information has been declared.

Please include your amended Competing Interests Statement within your cover letter. We will change the online submission form on your behalf.

Please know it is PLOS ONE policy for corresponding authors to declare, on behalf of all authors, all potential competing interests for the purposes of transparency. PLOS defines a competing interest as anything that interferes with, or could reasonably be perceived as interfering with, the full and objective presentation, peer review, editorial decision-making, or publication of research or non-research articles submitted to one of the journals. Competing interests can be financial or non-financial, professional, or personal. Competing interests can arise in relationship to an organization or another person. Please follow this link to our website for more details on competing interests: http://journals.plos.org/plosone/s/competing-interests

Additional Editor Comments (if provided):

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Yes

Reviewer #2: Yes

**********

2. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: Yes

Reviewer #2: Yes

**********

3. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: No

**********

4. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: Yes

**********

5. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: This paper addresses an important issue for health care systems in high income countries. It is analytically sophisticated and might be more consumable if the authors provided typical qualitative personas who might characterize the 4 classes that they identify. One also wonders about the heterogeneity within classes and whether a clinically meaningful approach to this segmentation process would require further breakdown within the 4 groups that emerge statistically in the analysis. The authors should make clear (as noted below) that segmentation suggests but does not establish clinically meaningful interventions.

132: are primary care visits associated with other incurred costs, or is this study useful only to improving management in the primary care setting.

285 ff: authors should be clear that presence of characteristics does not clarify the causal linkages with high utilization or the value of suggested interventions.

Reviewer #2: General comments

The paper offers a straightforward analysis of great relevance for clinical practice and for public policy. What seems like a simple exercise at first – the classification of high-needs, high costs patients (HNHC) into subgroups – resulted in a rigorous, yet interesting paper. On their own, these results may provide important insights and nuance into the understanding of HNHC, for example, that not all these patients are elderly. But they may also be an important starting point for additional qualitative work that could result into useful clinical guidelines. The paper was well-written and I enjoyed reading it. Having said that, some specific points in the methodology and analysis need revisions in order to improve the credibility of the study and strengths of the argument.

Data sources

• 104-105: Practices were invited to provide extract and provide individual-level data, and 48.5% of them responded. It would be important to add a comparison of the characteristics of the respondents and non-respondents. Is it likely that there is any bias in the pattern of responses that would affect the results?

• 108-109: Were the socioeconomic data individually linked to socioeconomic data? Please clarify.

Participants

• 131-132: Even though the study places considerable emphasis on high-cost patients, there is not data on costs generated by this group. Is there any cost data available? Otherwise, I strongly recommend re-labelling the group to high-needs, high-utilisation.

Data analysis

• 180-182: The process for elimination of variables is not transparent and could be interpreted as arbitrary. It would have been good to provide a list of variables that were not considered of clinical significance or had lower added clinical value.

Latent class analysis

• 202-203: While the cut-off of a minimum of 10% established by the authors for each class was clearly spell-out, it is very unlikely that this was a "a priori" design decision. Therefore, the decision to use 4 instead of 5 classes is not very well substantiated. It would be useful to see a comparison of a 5-class model. One way to do this would be to replace the current 1, which does not add any relevant information when compared to the tables for a comparative figure between a 4 and a 5 class model. Given that the main purpose of the paper is to conduct a classification, the argument needs to be stronger.

• 213-214: Are there any other possible living arrangements in class 2? Is there any variable for household size to confirm that these people were indeed living alone?

Table 1 and choice of variables

• The gender distribution of the sample is biased towards females. Why is that? Are females more likely to be high-needs, high-cost patients than men, controlling for age? Some discussion about that should be added to the text. Is it a problem of biased sample?

• How was the cut-off for children living at home decided? Is there any distinction between children and grandchildren in the data?

• For the prevalence of 28 chronic conditions, what is the criteria for inclusion? Are all of these diagnostics current? Do they include past diagnosis, and if so, for how long?

• While several proxies were used for socioeconomic status, some of the most common ones were not used, namely education and income levels. Within the selected variables, there could still be substantial variation. Why weren't education and income used?

**********

6. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: No

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files to be viewed.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email us at figures@plos.org. Please note that Supporting Information files do not need this step.

29 Nov 2019

Dear Dr. Heber, Maastricht, November 26, 2019

We are very grateful to be offered the opportunity to revise and resubmit our manuscript entitled: “Identifying subgroups of high-need, high-cost, chronically ill patients in primary care: A latent class analysis” for publication in PLoS ONE.

We appreciate the helpful feedback that was provided and believe that these comments were relevant and useful to improve our manuscript. Please find below our responses to the comments and reference to the changes we made in the manuscript. The changes in the revised manuscript are shown by ‘Track Changes’.

As requested, we would like to expand the ‘Competing interests’ statement: “AE is currently serving as an Academic Editor for PLOS ONE. Furthermore, our funding source (i.e. primary care group ‘HZD’) is a commercial source. This does not alter our adherence to PLOS ONE policies on sharing data and materials.”

We sincerely think that we have improved the quality of our paper and we hope that you now consider it for publication in PLoS ONE.

With kind regards, also on behalf of the co-authors,

Rowan Smeets, MSc

Maastricht University

Faculty of Health, Medicine and Life Sciences

Care and Public Health Research Institute (CAPHRI)

Department of Health Services Research

P.O. Box 616

6200 MD Maastricht, the Netherlands

T: +31(0)43 3881711 / F: +31(0)43 3884162

rowan.smeets@maastrichtuniversity.nl

Response to comments of reviewer #1

We thank the reviewer for the compliments and all the valuable comments provided on the original version of the manuscript. Below we indicate how we handled these comments. In the revised manuscript, all changes to the text are shown by ‘Track Changes’.

Comment: This paper addresses an important issue for health care systems in high income countries. It is analytically sophisticated and might be more consumable if the authors provided typical qualitative personas who might characterize the 4 classes that they identify. One also wonders about the heterogeneity within classes and whether a clinically meaningful approach to this segmentation process would require further breakdown within the 4 groups that emerge statistically in the analysis. The authors should make clear (as noted below) that segmentation suggests but does not establish clinically meaningful interventions.

Our response: As suggested by the reviewer, we added typical qualitative personas who characterize the four classes as supporting information (with reference to this supporting information, S1 File, in the revised results section) to make the classes more consumable. These qualitative personas include information about the most distinct patient characteristics between the classes (i.e. age, household position, source of income). In addition, information about highly prevalent chronic conditions was included in the personas. For example, the following persona of class 2 ‘older adults living alone’ was provided: “Mrs. Williams is 79 years old and living alone. Her husband has passed away five years ago. For some time now, Mrs. Williams has to deal with visual disorders and osteoarthritis. In addition, she has been suffering from diabetes for a long time.” The segmentation approach presented in the current study is particularly of clinical relevance as it can guide towards (rather than establish) more tailored interventions, in close interaction with the patient. To describe this, we have added the following to the revised discussion: “As such, this segmentation approach can serve as a starting point for more biopsychosocial attention and can inform the discussion of tailored interventions with the patient (56). However, the individual consultation is still key to assess personal needs and preferences with a patient during a consultation, and agree on an individual treatment course.” (page 21, lines 316-320)

Comment: 132: are primary care visits associated with other incurred costs, or is this study useful only to improving management in the primary care setting.

Our response: Our aim was to identify classes of chronically ill patients who can be considered HNHC in primary care, based on their utilization of GP consultations. It was beyond of the scope of the current study to investigate to what degree GP consultations are associated with other incurred costs. This means that the insights retrieved from this study most directly inform management in the primary care setting. Nevertheless, we acknowledge that the association between GP consultations and other incurred costs refers to an interesting topic for future scientific research. Therefore, we added the following to the revised discussion: “In addition, the current study has focused on HNHC patients in primary care, which is widely considered the most suitable medical home for chronically ill patients (57). Although as a result, our findings are mainly useful for improvement of primary care management, there is some evidence that patients with a disproportionately high use of primary care resources also account for significantly high(er) costs in specialist care” (58, 59) (pages 21-22, lines 323-328)

Comment: 285 ff: authors should be clear that presence of characteristics does not clarify the causal linkages with high utilization or the value of suggested interventions.

Our response: We agree with the reviewer that the aim of our analysis should be clearer in the manuscript. Our aim was not to identify relations between a (set of) independent variable(s) and a dependent variable, which are then assumed to hold across all people. Rather than using such a variable-oriented analytic approach, we used an approach more befitting population segmentation purposes, i.e. latent class analysis (LCA). LCA is a person-oriented analytic approach that emphasizes the patient as a whole. LCA aims to identify classes of individuals with similar patterns of personal factors relevant to, in this study, health care use. In selecting personal factors to include in the LCA, we chose a range of 41 characteristics that have been identified as relevant in relation to health care utilization in previous studies.

To clarify this, the following was added to the revised methods section: “Furthermore, LCA is a person-oriented analysis technique (37) which aims to identify classes of individuals with similar patterns of, in the current study, (correlated) personal factors relevant to health care utilization.” (page 9, lines 160-162) Furthermore, we have added the following to the revised methods section: “Forty-one patient characteristics were initially used as potential indicators of heterogeneity in HNHC patients’ needs in the LCA. These characteristics were included based on scientific studies describing these characteristics as relevant in relation to (high) care utilization (12, 36).” (page 8, lines 137-138) Moreover, we have added the following to the revised discussion: “This view can suggest potentially relevant goals, interventions, and professionals (within primary care and in cooperation with other disciplines), which can be further discussed in a shared decision-making process with the patient. Such an approach can be inspired by ‘The Bridges to Health Model’ (55) which aims to systematically connect priority concerns, major components of health care, and goals for health care within identified population segments (56).” (page 21, lines 309-314)

Response to comments of reviewer #2

We thank the reviewer for the compliments and all the valuable comments provided on the original version of the manuscript. Below we indicate how we handled these comments. In the revised manuscript, all changes to the text are shown by ‘Track Changes’.

Comment: 104-105: Practices were invited to provide extract and provide individual-level data, and 48.5% of them responded. It would be important to add a comparison of the characteristics of the respondents and non-respondents. Is it likely that there is any bias in the pattern of responses that would affect the results?

Our response: Although we do not have individual patient data of the non-responding practices, we have compared patient characteristics (i.e. sex, age, household position, and source of income) of responding practices with the general population from this northern region of the Netherlands (that is covered by the primary care group). The (aggregated) data from the general population from this northern region was accessed via the open access database of Statistics Netherlands (i.e. ‘StatLine’). This comparison showed that the participating practices are comparable to the general population in the specific region. To clarify this, we have added the following to the revised discussion: “First, individual level data of the non-participating practices were not available in this study. This hampered a direct comparison of participating practices (n= 63; 48.5% of care group) with non-participating practices in order to assess representativeness of the sample. However, particular patient characteristics (i.e. sex, age, household position, and source of income) of the sample were compared to the patient characteristics of the general population in the northern region of the Netherlands that is covered by the primary care group. This comparison showed that the sample is largely similar in patient characteristics to the general population. For example, 50.8% of the sample is female; 50.5% of the general population is female, 20.1% of the sample receives pension benefits; 22.1% of the general population receives pension benefits.” (page 22, lines 337-346)

Comment: 108-109: Were the socioeconomic data individually linked to socioeconomic data? Please clarify.

Our response: All extracted data were individually linked. To more explicitly describe this, we have added the following to revised methods section: “Furthermore, the EHR data were linked on the individual patient level to socioeconomic data (e.g., source of income) and health care claims data (e.g., pharmaceutical costs).” (page 6, lines 107-108)

Comment: 131-132: Even though the study places considerable emphasis on high-cost patients, there is not data on costs generated by this group. Is there any cost data available? Otherwise, I strongly recommend re-labelling the group to high-needs, high-utilisation.

Our response: We use the terminology of ‘high-need, high-cost’ (HNHC) as the weighting factors that are assigned to the different (primary care) consultation types are related to costs (as we described in the original version of the manuscript under ‘participants’). As such, the population can be considered ‘high-need’ as well as ‘high-cost’, but primarily within our research setting, i.e. primary care. To more explicitly describe this, we added the following to the revised methods section: “As the weighting factors based on time investment are related to costs (35), the patients selected for this study can be considered high-need, high-cost in primary care.” (page 7, lines 133-134)

Comment: 180-182: The process for elimination of variables is not transparent and could be interpreted as arbitrary. It would have been good to provide a list of variables that were not considered of clinical significance or had lower added clinical value.

Our response: To identify clinically relevant subgroups within the HNHC population, we initially employed latent class analysis by using 41 patient characteristics that were considered relevant in relation to (high) care utilization. These 41 characteristics are displayed in table 1. In Table 3, the nine variables included in the final model are shown. This means that 32 variables were excluded (all due to statistical reasons). To clarify this, we added the following to the revised results section: “This means that the following variables were excluded in the final LCA due to less statistical relevance: age of children living at parental home, number of chronic conditions, prevalence of 28 chronic conditions, paid interest over debts, GP care utilization on baseline.” (page 11-12, lines 216-218)

Comment: 202-203: While the cut-off of a minimum of 10% established by the authors for each class was clearly spell-out, it is very unlikely that this was a "a priori" design decision. Therefore, the decision to use 4 instead of 5 classes is not very well substantiated. It would be useful to see a comparison of a 5-class model. One way to do this would be to replace the current 1, which does not add any relevant information when compared to the tables for a comparative figure between a 4 and a 5 class model. Given that the main purpose of the paper is to conduct a classification, the argument needs to be stronger.

Our response: The cut-off of 10% was in fact an important “a priori” decision criterion, because the model also has to be feasible for smaller GP practices that want to provide tailored care, i.e. practices with <2000 patients and <120 HNHC chronically ill patients. Nevertheless, statistical robustness and clinical relevance were firstly assessed and regarded as most important criteria for model choice. To our best knowledge, it is quite uncommon (if the study’s aim is to identify rather than to compare a model) to present a fully detailed (visual) comparison of different models. After all, decision criteria are commonly defined “a priori” and inform model choice. Usually, only a comparison of statistical indicators for models with different number of classes is given (see Table 2 in the original manuscript) and a general substantive argument to underpin the choice of the final model. To make the substantive argument for the choice of the 4-class model stronger, we added the following to the revised results section: “More specifically, a 5-class model largely maintained three of the four classes of the 4-class model and subdivided the fourth and smallest class of the 4-class model into two smaller classes which were relatively indistinct from each other.” (page 11, lines 210-212)

Comment: 213-214: Are there any other possible living arrangements in class 2? Is there any variable for household size to confirm that these people were indeed living alone?

Our response: In class 2, the household position of ‘single adult’ was most dominantly present (0.917), followed by ‘member of a collective household’ (0.08), ‘single parent’ (0.002), and ‘other’ (0.001). There are no other living arrangements in class 2 (see ‘household position’ in Table 3). In our dataset, we have one variable for household size (which was not included in the analysis as it correlates with household position and number of people with an individual income in a household), which confirms that the majority of this class (i.e. 99%) is part of a household with only one member.

Comment: The gender distribution of the sample is biased towards females. Why is that? Are females more likely to be high-needs, high-cost patients than men, controlling for age? Some discussion about that should be added to the text. Is it a problem of biased sample?

Our response: In the current study, we have used a person-oriented rather than a variable-oriented analysis to seek for subgroups of individuals based on their patterns of patient characteristics. This implies that we did not focus on the relation between (sets of) individual characteristics, with corrections for possibly confounding variables. Whereas the included sample is, compared to the general population, not biased (see response to comment about participating practices), the included HNHC population shows an unbalanced distribution of sex. Several explanations for this unbalanced distribution in sex were added to the revised discussion: “The current study indicates that the sex distribution within the HNHC population, as well as in three of the four identified subgroups, is unbalanced: more than 64% of the HNHC population is female. In the current person-oriented analysis, unlike in a variable-oriented analysis, there is no assessment of relations between variables including corrections for confounders. Rather, the current analysis has focused on identifying subgroups based on patterns of variables within individual patients. One possible explanation for the unbalanced population in terms of sex is that women typically get older and, as a result, are overrepresented among the older aged HNHC patients compared to men. In addition, scientific studies have found that women have significantly higher consultation rates compared to men, but particularly during working years (46, 47).” (pages 19-20, lines 272-281)

Comment: How was the cut-off for children living at home decided? Is there any distinction between children and grandchildren in the data?

Our response: In response to this comment, we added the following text to the revised methods section: “Age of children living at parental home (≤12 years, >12 years (i.e. the age that they generally leave elementary school), no children living at parental home).” (page 8, lines 142-143) Furthermore, the extracted data (amongst others the variable of ‘Age of children living at home’) only includes information about children (living at home) with regards to household situation; there is no information available on grandchildren (living at home). To make this more clear, we have changed the name of the variable “Age of children living at home” into “Age of children living at parental home”.

Comment: For the prevalence of 28 chronic conditions, what is the criteria for inclusion? Are all of these diagnostics current? Do they include past diagnosis, and if so, for how long?

Our response: The prevalence of the chronic conditions can be considered the current prevalence/diagnosis as it is based on recent GP care use for the particular condition. To describe this, we added the following to the revised methods section: “Biomedical characteristics were also measured at baseline and included patients’ chronic disease diagnoses based on GP care use related to the chronic disease in the 1.5 years before baseline.” (page 8, lines 143-145)

Comment: While several proxies were used for socioeconomic status, some of the most common ones were not used, namely education and income levels. Within the selected variables, there could still be substantial variation. Why weren't education and income used?

Our response: Educational level was not used as this variable was not available to us on an individual patient level for every (or the majority of) patient(s) covered by the primary care group. Statistics Netherlands only has information on educational level for a sample of the region covered by the primary care group. In addition, individual income was not used (as a continuous variable) as the ‘source of income’ was regarded more informative than the amount of individual income. Therefore, we preferred the inclusion of ‘source of income’ over individual income as a continuous variable.

Submitted filename: Response to reviewers_HNHC subgroups_29-11-2019.docx

Click here for additional data file.

8 Jan 2020

Identifying subgroups of high-need, high-cost, chronically ill patients in primary care: A latent class analysis

PONE-D-19-25812R1

Dear Dr. Smeets,

We are pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it complies with all outstanding technical requirements.

Within one week, you will receive an e-mail containing information on the amendments required prior to publication. When all required modifications have been addressed, you will receive a formal acceptance letter and your manuscript will proceed to our production department and be scheduled for publication.

Shortly after the formal acceptance letter is sent, an invoice for payment will follow. To ensure an efficient production and billing process, please log into Editorial Manager at https://www.editorialmanager.com/pone/, click the "Update My Information" link at the top of the page, and update your user information. If you have any billing related questions, please contact our Author Billing department directly at authorbilling@plos.org.

If your institution or institutions have a press office, please notify them about your upcoming paper to enable them to help maximize its impact. If they will be preparing press materials for this manuscript, you must inform our press team as soon as possible and no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

With kind regards,

Bruno Pereira Nunes, Ph.D.

Academic Editor

PLOS ONE

Additional Editor Comments (optional):

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation.

Reviewer #1: All comments have been addressed

Reviewer #2: All comments have been addressed

**********

2. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Yes

Reviewer #2: (No Response)

**********

3. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: Yes

Reviewer #2: (No Response)

**********

4. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: (No Response)

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: (No Response)

**********

6. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: I have no additional comments for the authors or for the editors. The authors have carefully considered and responded to all comments.

Reviewer #2: This is a revised version submitted by the authors in response to a previous round of review. The authors have satisfactorily addressed the comments I raised.

**********

7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: No

13 Jan 2020

PONE-D-19-25812R1

Identifying subgroups of high-need, high-cost, chronically ill patients in primary care: A latent class analysis

Dear Dr. Smeets:

I am pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department.

If your institution or institutions have a press office, please notify them about your upcoming paper at this point, to enable them to help maximize its impact. If they will be preparing press materials for this manuscript, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact onepress@plos.org.

For any other questions or concerns, please email plosone@plos.org.

Thank you for submitting your work to PLOS ONE.

With kind regards,

PLOS ONE Editorial Office Staff

on behalf of

Dr. Bruno Pereira Nunes

Academic Editor

PLOS ONE