Preferences for hospital in the home for COVID-19: a discrete choice experiment study

Abstract

Objective

This study aimed to assess Tehran residents’ preferences for COVID-19 hospital in the home, using a discrete choice experiment (DCE), to support evidence-based policy-making for future health crises.

Methods

Nine hundred twenty-nine Tehran residents aged ≥ 18 years using a multistage sampling approach were surveyed. Data was collected via a structured questionnaire. Attributes (price, physician specialty, emergency contact availability, and caregiver’s qualification) and levels were identified through literature review, and consultation with the experts. Data were analyzed by latent class conditional logit model, with class membership examined via a logit model. Marginal willingness-to-pay (MWTP) were calculated, and analyses were performed in Stata version 14 with 95% confidence interval.

Results

A total of 929 participants (mean age = 36.8, SD = 10.8; 52.7% male) were included. Two latent classes were identified. In Class 1, participants were more price-sensitive (CE = − 0.02, p < 0.001), with a strong preference for emergency contact availability (CE = 5.69, p < 0.001) and combined physician teams (CE = 3.89, p < 0.001). In Class 2, price sensitivity was weaker (CE = − 0.002, p < 0.001), with lower but significant WTP values (111.95 USD for combined physician teams; 56.59 USD for emergency contact). Logistic regression showed that older age (p < 0.01), higher education (OR = 0.32, p < 0.001), very good health status (OR = 0.46, p = 0.001), COVID-19 hospitalization history (OR = 0.47, p < 0.001), and having social security/ health/ other basic insurance (p < 0.05) reduced the likelihood of class 2 membership, whereas larger family size (OR = 1.23, p = 0.003) and low income (p < 0.01) increased it; being married had a borderline significant positive association (OR = 1.52, p = 0.05).

Conclusion

While both groups were price-averse, decisions were mainly driven by emergency support, clinical quality (physician specialty and caregiver’s qualification). These findings underscore the need for patient-centered hospital in the home care models that integrate affordability with responsive and high-quality services.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13561-026-00798-w.

Introduction

COVID-19, declared a pandemic on March 11, 2020, led to social distancing strategies and an expansion of critical care capacity [1]; however, workforce limitations and technical and infrastructural challenges prevented a full response [2]. With an estimated ICU admission rate of approximately 5%, demand quickly exceeded global hospital capacity, resulting in severe shortages of resources and essential services [3]. Beyond the shortage of trained specialists for safe ventilator use and high staff mortality, limited hospital and ICU beds posed major challenges [4, 5]. The COVID-19 pandemic highlighted the relative scarcity of hospital resources, with one of the main factors contributing to mortality being the limited capacity, particularly intensive care unit (ICU) beds [6, 7]. Concurrently, the World Health Organization recommended home-based care as a primary model to meet the medical needs of patients and vulnerable populations during the pandemic [4].

Given the mutating nature of SARS-CoV-2 and the potential emergence of new variants with higher transmissibility or severity, health systems must prepare for possible future epidemics Recurrent infection waves, limited effectiveness of some public health measures, and uncertainty regarding vaccine efficacy against emerging variants further highlight this need [8–10], the limited effectiveness of public health measures [11], and the potential impact of new variants on vaccine efficacy, diagnostics, and treatments [12, 13]—along with the fact that vaccines cannot fully prevent infection [14]—health systems must prepare for possible future epidemics [15]. In this context, home-based care—defined by WHO as a system for caring for patients at home to prevent or delay the need for long-term or acute medical care—can help manage hospital saturation [4]. The primary mission of these services is to maintain and restore patient independence, enabling individuals to meet their health needs at home with the support of healthcare providers while relying on their own abilities [16]. Evidence shows that home healthcare could be an important part of the health system, providing high-quality and cost-effective care [17] that helps vulnerable individuals maintain independence [18, 19] and improve their quality of life [20]. Although caring for infected individuals at home increases the risk of transmission within the household, isolating patients can play a crucial role in breaking the chain of infections. Decisions about home isolation and care depend on clinical assessment, home environment evaluation, and the ability to monitor clinical progress [21]. Home care benefits patients by preserving privacy, dignity, and comfort, and may be considered in cases of mild illness, low risk of rapid deterioration, the ability for family follow-up, or limited/unsafe hospital capacity. Patients with mild symptoms and no high-risk comorbidities can be cared for at home [21, 22]. Practically, diagnostic and care facilities are provided according to what is feasible in a typical household, with professional nurses and primary care physicians forming the core of these services [23].

In Iran, 704 licensed home healthcare centers operate across 30 provinces, providing in-home medical services in conjunction with some university-affiliated hospitals [24]. Considering the simultaneous presence of this capacity alongside structural challenges in the health system (human resource shortages, limited hospital beds, and constrained resources), expanding community-based home care can support the management of care needs [25]. Therefore, conducting studies that assist policymakers in decisions regarding the use of home-based inpatient care, sometimes called hospital in the home, during periods of limited hospital resources is important. One such approach is the Discrete Choice Experiment (DCE), which is widely used to identify priorities in health and healthcare. DCEs are often used to value outcomes, examine trade-offs between health and non-health outcomes, and develop priority-setting frameworks [26, 27]. In this method, potential interventions or products are described by attributes, each presented at one level chosen from a finite set of possible levels that vary over a plausible and policy relevant range. These are then combined using designed experiments and presented as sets of hypothetical options for respondents to choose from. Ultimately, preferences based on attributes and levels are extracted [28].

Previous discrete choice experiment studies conducted in other countries have examined public preferences for home-based healthcare and hospital in the home services. These studies have shown that attributes such as provider expertise, continuity of care, emergency support availability, quality of care, and out-of-pocket costs play important roles in individuals’ decisions regarding home-based healthcare services [29–32]. In addition, prior research demonstrated substantial heterogeneity in preferences according to demographic, socioeconomic, and health-related characteristics [29]. However, evidence remains limited regarding preferences for COVID-19-related hospital in the home services, particularly in low- and middle-income countries and within the Iranian healthcare context.

To the best of our knowledge, no published study has examined Iranian residents’ preferences for COVID-19 hospital in the home services using a DCE approach. Understanding these preferences is important given the potential role of home-based care in managing future infectious disease outbreaks and periods of hospital resource constraints. Therefore, this study aimed to investigate Tehran residents’ preferences regarding COVID-19 hospital in the home services and to identify the relative importance of different service attributes and sources of preference heterogeneity. Specifically, the study sought to answer the following research question: which characteristics of home-based COVID-19 care services are most valued by Tehran residents, and how do these preferences vary across population groups? The findings are expected to inform policymakers in designing acceptable, efficient, and patient-centered home-care strategies during future public health crises.

Method

Study design and population

This descriptive-analytical, cross-sectional study was conducted from May to November 2023. The target population consisted of all residents of Tehran city, Iran, aged over 18 years.

Sample size calculation and sampling

The sample size was determined using the formula N > 500c/(t x a), where t is the number of choices sets each respondent answers, a is the number of alternatives per choice set, and c is the maximum number of levels among the attributes [33, 34]. Based on our design (t = 9, a = 2, c = 3), a minimum of 84 respondents was required. To ensure robustness, 1,000 residents were actually sampled. A multistage sampling approach was applied. In the first stage, Tehran’s 22 municipal districts were stratified according to socioeconomic status into four categories: less developed, semi-privileged, deprived, and developed/highly developed. Several districts were then randomly selected from each socioeconomic category, resulting in a total of 10 districts selected for data collection. The total sample size was proportionally allocated according to the population size of the selected districts. In the second stage, urban community health centers were randomly selected within each selected district. The required sample size for each selected center was then determined proportionate to the population covered by that center relative to the total population covered within the district. In the final stage, eligible individuals attending the selected centers were recruited using a convenience sampling approach based on willingness to participate.

Of the 1,000 invited participants, 929 completed valid questionnaires after excluding 71 inconsistent responses where a less attractive alternative was chosen as better (often called a dominance test). No evidence of non-trading behavior (e.g., selecting the same alternative across all choice tasks) was observed among the 929 participants who completed valid questionnaires.

Data collection and questionnaire

Data was collected using a structured questionnaire consisting of two sections: (1) demographic and socio-economic characteristics and (2) DCE tasks. The questionnaire was self-administered via a web-based platform (Porsline). Participants completed the survey by scanning a QR code at selected health centers using their mobile devices after providing informed consent. No commercial survey panel or external sampling provider was used. Regarding completion time, the online platform (Porsline) automatically recorded response durations. No questionnaire was completed in less than 10 min. All respondents who completed the survey did so within a reasonable timeframe (10–25 min), and no one was excluded based on response speed.

In the first section, demographic, health and socio-economic variables (age, sex, marital status, family size, being household head, owning a house, educational level, employment status, monthly income level, monthly expenditures level, health status, having underlying diseases, having basic health insurance, having supplementary health insurance, history of COVID-19 infection, history of hospitalization due to COVID-19, history of using home care services, satisfaction from home care service) were collected. The second section of questionnaire consists of the choice tasks. Each choice task included two hypothetical alternatives describing home-based COVID-19 care services, characterized by four attributes: price, physician specialty, emergency contact availability, and caregiver qualification. Each respondent completed one block of nine choice sets.

DCE design

The development of the DCE followed three steps: (1) identification of attributes and levels (2), construction of choice tasks, and (3) pilot testing.

Attribute selection

Initially, a focused review of the literature was conducted, followed by consultations with academic experts and professionals familiar with hospital-in-the-home services. The initial list of attributes was reviewed and refined by the research team and an expert panel. As a result, 16 candidate attributes were identified. Because inclusion of all identified attributes was not feasible within the DCE design, an attribute rating approach was applied to prioritize attributes according to their perceived importance. Attribute screening was then conducted among 15 citizens and 15 clinical experts, who rated each attribute using a 5-point Likert scale. The four attributes with the highest mean importance scores were selected for inclusion in the DCE. Appropriate attribute levels were subsequently determined through expert consultation. The full list of candidate attributes and their rankings is provided in Appendix 1.

The final attributes and levels were:

1. Home-based hospitalization price (per day): free, 67.12 USD, 111.39 USD

2. Physician specialty: general practitioner, specialist, both general practitioner and specialist

3. Emergency contact availability: yes, no

4. Caregiver qualification: non-registered nurse, registered nurse

Expert consultation and review of prevailing private-sector tariffs in Tehran were used to define attribute levels, particularly for price. These levels were designed to reflect realistic market conditions and plausible policy scenarios. The zero-price level was included as a hypothetical policy scenario representing fully subsidized home-based care rather than a market-based market price.

Experimental design

A fractional factorial design based on a Kuhfeld [35] orthogonal array was used to construct the choice tasks. A full factorial design would have generated 630 possible combinations, which was not feasible. The final design consisted of 36 choice sets, which were blocked into four versions of nine choice tasks each. Each respondent completed one block. The blocks were constructed to have attribute level overlap so that respondents could not always chose based on one attribute. The versions were chosen to ensure that each attribute appeared at the same level in at least two choice sets in each version. The choice sets were chosen to allow for efficient estimation of attribute effects. This approach to design construction follows Street and Burgess (2007). The design achieved a D-efficiency of 74.93% under a null prior. No opt-out option was included to ensure forced trade-offs between alternatives. A choice set in which one option was clearly better than the other option, a dominance test, was embedded within the questionnaire to assess respondent engagement.

Pilot study

A pilot study was conducted with 40 participants to assess comprehension of attributes, clarity of choice tasks, and overall feasibility. Minor revisions were made based on feedback. In addition, a conditional logit model was applied, and all coefficients had the expected signs. Since the questionnaire was developed based on the consultations with the experts, its content validity was confirmed. Moreover, the use of orthogonal design and other statistical methods in generating the choice tasks ensured that the instrument possessed adequate reliability and construct validity.

Utility specification

The random utility model was specified as follows:

where represents the utility of alternative j for individual i. Categorical variables were dummy-coded and price was treated as a continuous variable.

Statistical analysis

Descriptive statistics (mean, standard deviation, and frequencies) were used to summarize participant characteristics.

A mixed logit model was first estimated to assess preference heterogeneity. Given the presence of heterogeneity, a latent class logit model was applied to identify unobserved subgroups with similar preference structures. Each respondent was assigned to the class with the highest posterior probability. To determine the optimal number of classes, models with two to six classes were estimated using Stata (lclogitml2). Models with three or more classes failed to converge despite multiple random starting values, likely due to model identification issues and class instability; therefore, the two-class model was selected based on model stability, convergence, and interpretability.

The marginal willingness-to-pay (MWTP) for each attribute was then calculated within each class using the formula:

To determine the variables related to class membership, posterior probabilities were used in a logit model.

All costs, originally collected in Iranian Rials, were converted to international dollars (USD) using the purchasing power parity (PPP) conversion factor (89391.81IRR per international dollar, year 2023) [36]. Data analysis was performed using Stata version 14 software.

Ethical approval

Prior to beginning the study, it was approved by the Hamadan University of Medical Sciences ethics committee (ethical code: IR.UMSHA.REC.1401.295). Participants provided their explicit written informed consent during the recruitment phase by agreeing to participate. Personal identifying information was not collected, and participant responses were anonymized prior to analysis.

Results

The study encompassed 929 participants with a mean age of 36.76 years (SD = 10.83) and an average family size of 3.63 (SD = 1.27). Males constituted 52.74% of the sample. Most participants were married (72.98%), and 43.81% were the household heads. About half of the participants owned their home (47.79%).

Education was led by bachelor’s degrees (31.75%), followed by 8–12 years of schooling (27.23%). PhD holders comprised 4.41%, while 0.34% had less than five years of formal education. Social security insurance covered 52.74% of participants, and less than half had supplementary insurance (41.66%). Only 12.49% reported prior use of home-care services, all of whom were satisfied. COVID-19 experiences showed 65.55% had infection without hospitalization, and 1.29% required hospitalization. Underlying health conditions were reported by 21.21%, and self-rated health was largely very good (46.93%).

Socioeconomic status showed that 37.76% of the participants had monthly incomes between 894.94 USD to 1678 USD, and most of them (35.52%) had monthly expenditures of 426.73 USD to 853.46 USD. Other participants’ characteristics are presented in Table 1.

Table 1.

Characteristics of the study participants (n = 929)

Variable		n (%) /mean (S.D)	Variable		n (%) /mean (S.D)
Age		36.76 ± 10.83	Employment	Yes	603 (64.91)
Family size		3.63 ± 1.27		No	326 (35.09)
Sex	Male	490 (52.74)	Basic health insurance	No insurance	86 (9.26)
	Female	439 (47.26)		Social security insurance	490 (52.74)
Marital status	Single (never married)	231 (24.86)		Health insurance	212 (22.82)
	Married	678 (72.98)		Armed forces insurance	92 (9.90)
	Divorced/Widow	20 (2.16)		Other	49 (5.27)
Household head	Yes	407 (43.81)	Supplementary health insurance	Yes	387 (41.66)
	No	522 (56.19)		No	542 (58.34)
Owning a house	Yes	444 (47.79)	History of getting home care services	Yes	116 (12.49)
	No	485 (52.21)		No	813 (87.51)
Education	< 5 years	31 (3.34)	Satisfaction of used home care services*	Yes	116 (100)
	8–12 years	253 (27.23)	History of COVID-19 infection	No history of COVID-19 infection	271 (29.17)
	Associate degree	109 (11.73)		Yes, hospitalized at governmental hospital	37 (3.98)
	Bachelor’s degree	295 (31.75)		Yes, hospitalized at private hospital	12 (1.29)
	Master’s degree	109 (11.73)		Yes, no hospitalization	609 (65.55)
	PhD	41 (4.41)	Monthly income (USD)	No income	178 (19.16)
	Seminary studies	7 (0.75)		<894.94	167 (17.98)
Having underlying diseases	Yes	197 (21.21)		894.94–1678	350 (37.67)
	No	732 (78.79)		1678-2796.68	169 (18.19)
Health status	Very good	208 (22.39)		> 2796.68	65 (7.00)
	Good	436 (46.93)	Monthly expenditures (USD)	< 559.33	222 (23.90)
	Moderate	253 (27.23)		559.33-1118.67	330 (35.52)
	Poor	27 (2.91)		1118.67–1678	247 (26.59)
	Very poor	5 (0.54)		> 1678	130 (13.99)

* Only respondents with prior home-care experience answered this item

The latent class model with two classes provided the best model fit. Based on the class probabilities 27% (n = 251) of all respondents was assigned to class 1 and 73% (n = 678) to class 2. For the two-class model, the average of the respondents’ maximum posterior class membership probabilities was 0.97 (SD = 0.07), ranging from 0.57 to 1.00. The vast majority of respondents (97.3%) had a maximum class membership probability ≥ 0.70, indicating a high degree of class separation (Table 2).

Table 2.

Results of the latent class conditional logit model and willingness to pay for COVID-19 home-based care

Variable	Class 1			Class 2
	CE	SE	95% CI	CE	SE	95% CI
Price	-0.02**	0.004	(-0.03, − 0.01)	-0.002**	0.0004	(-0.003, -0.002)
Physician specialty (Ref: General)
Specialist	-0.11	0.17	(-0.44, 0.21)	0.19**	0.45	(0.10, 0.28)
General and specialist	3.89**	0.59	(2.74, 5.04)	0.28**	0.45	(0.20, 0.37)
Emergency contact (Ref: No)
Yes	5.69**	0.84	(4.04, 7.339)	0.14**	0.04	(0.07, 0.22)
Caregiver qualification (Ref: Non-Registered nurse)
Registered nurse	1.94**	0.32	(1.30, 2.57)	0.09**	0.03	(0.03, 0.15)
Shared
Constant	-0.98**	0.09	(-1.17, -0.80)
Log likelihood= -4982.78, CAIC = 10051.74, AIC = 9987.57, BIC = 10040.74, N = 929

Willingness to pay (USD)	Class 1			Class 2
	WTP	SE	95% CI	WTP	SE	95% CI
Physician specialty (Ref: General)
Specialist	-5.08	6.90	(-18.60, 8.44)	75.69**	20.34	(35.82, 115.55)
General and specialist	172.71**	18.59	(136.27, 209.15)	111.95**	22.69	(67.48, 156.43)
Emergency contact (Ref: No)
Yes	252.89**	23.06	(207.68, 298.09)	56.59**	15.79	(25.65, 87.53)
Caregiver qualification (Ref: Non-Registered nurse)
Registered nurse	86.06**	6.56	(73.21, 98.92)	34.96*	12.88	(9.71, 60.21)

CE Coefficient Estimate, SD Standard Error, CI Confidence Interval

* P <0.01, ** P <0.001

It should be noted that the estimated coefficients represent relative utility weights in the latent class model and are scale-dependent; therefore, their magnitude should be interpreted comparatively rather than in absolute terms. The results of the latent class regression model revealed substantial differences in price sensitivity between the two identified classes. In class 1, the negative and statistically significant coefficient for price (–0.022, p < 0.001) indicated a marked decrease in the likelihood of choosing the service as cost increased. In contrast, in class 2, the effect was much weaker, with a small negative coefficient (–0.002, p < 0.001), suggesting only minimal price sensitivity.

Regarding physician specialty (with general practitioners as the reference category), in class 1 the presence of a specialist physician alone had no significant effect (–0.114, p = 0.493), whereas a combination of general practitioner and specialist physicians exerted a strong positive influence on service preference (3.887, p < 0.001). In class 2, both the specialist physician option (0.192, p < 0.001) and the combined physician team (0.284, p < 0.001) had significant positive effects, though their magnitudes were smaller compared to those of class 1.

The availability of an emergency contact had a significant positive effect in both classes, but the magnitude was substantially greater in class 1 (5.692, p < 0.001) than in class 2 (0.143, p < 0.001). Similarly, having a registered nurse as the caregiver (versus a non-registered nurse) significantly increased preference in both classes; however, the effect was much stronger in class 1 (1.937, p < 0.001) than in class 2 (0.089, p < 0.01). Overall, these findings suggest that class 1 appears to exhibit higher price sensitivity and places much greater emphasis on the safety and quality of home-care services, whereas class 2 demonstrates lower price sensitivity and smaller differences in the valuation of service attributes.

The analysis of WTP for hospital in the home COVID-19 care services highlighted clear differences between the two latent classes. However, it should be noted that the negative willingness-to-pay estimates may arise due to the ratio-based calculation (attribute coefficient divided by the price coefficient). When the corresponding attribute coefficient is small or statistically insignificant, WTP estimates may become negative, reflecting weak or non-significant preferences rather than a true negative valuation.

In class 1, respondents exhibited the highest WTP for the combination of a general practitioner and a specialist, estimated at 172.71 USD (95% CI: (136.27, 209.15), p < 0.001), indicating a strong preference for this service configuration. The presence of a specialist physician alone did not significantly affect WTP (–5.08 USD, 95% CI: (–18.60, 8.44), p = 0.46). Additionally, the availability of an emergency contact substantially increased WTP by 252.89 USD (95% CI: (207.68, 298.09), p < 0.001), and having a registered nurse as the caregiver raised WTP by 86.06 USD (95% CI: (73.21, 98.92), p < 0.001).

In Class 2, the WTP values were generally lower but still significant for all attributes. A specialist physician alone increased WTP by 75.69 USD (95% CI: (35.82, 115.55), p < 0.001), while a combined general practitioner and specialist team raised WTP by 111.95 USD (95% CI: (67.47, 156.43), p < 0.001). The presence of an emergency contact contributed an additional 56.59 USD (95% CI: 25.65–87.53, p < 0.001), and having a registered nurse as the caregiver increased WTP by 34.96 USD (95% CI: (9.71, 60.207), p < 0.01).

Overall, these results suggest that class 1 respondents place a particularly high monetary value on service attributes related to quality and reliability, whereas class 2 respondents exhibit lower but still meaningful WTP, reflecting heterogeneity in the perceived value of home-based care services.

The results of the logistic regression model identifying determinants of latent class membership are presented in Table 3. Age was negatively associated with the likelihood of belonging to class 2, with respondents aged 34–48 showing lower odds (OR = 0.52, p = 0.001) and those older than 48 years also less likely to belong to class 2 (OR = 0.44, p = 0.004) compared to respondents aged ≤ 33. Marital status increased the probability of class 2 membership, with married individuals more likely to belong to class 2 than single, widowed, or divorced respondents (OR = 1.521, p = 0.05). This result was at the threshold of statistical significance.

Table 3.

Results of logit regression for determinants of class 2 membership

Variable	CE	OR	SE	95% CI
Constant	2.22	6.44*	3.62	(2.14, 19.39)
Sex (Ref: Male)
Female	0.38	1.47	0.38	(0.88, 2.44)
Age (Ref: ≤33)
33 < age ≤ 48	-0.65	0.52*	0.11	(0.35, 0.78)
48<	-0.82	0.44*	0.13	(0.25, 0.78)
Head of household (Ref: No)
Yes	0.22	1.24	0.34	(0.72, 2.14)
Owner of house (Ref: No)
Yes	-0.13	0.88	0.15	(0.62, 1.24)
Marital status (Ref: Single/Widow/Divorced)
Married	0.42	1.52	0.32	(1.00, 2.23)
Health status (Ref: Bad/Very bad/Moderate)
Good	0.04	1.05	0.21	(0.70, 1.56)
Very good	-0.78	0.46*	0.11	(0.29, 0.72)
Underlying disease (Ref: No)
Yes	0.12	1.12	0.25	(0.17, 1.75)
Insurance (Ref: No insurance)
Social security insurance	-0.99	0.37*	0.15	(0.17, 0.80)
Health insurance	-0.90	0.41*	0.17	(0.17, 0.94)
Armed forces insurance	-0.52	0.59	0.22	(0.29, 1.22)
Other	-1.41	0.24*	0.11	(0.10, 0.61)
Supplementary insurance (Ref: No)
Yes	-0.21	1.24	0.22	(0.87, 1.76)
Employment (Ref: Unemployed)
Employed	-0.32	0.73	0.16	(0.47, 1.3)
Education (Ref: Less than college)
Associate or bachelor’s degree	-1.15	0.32**	0.06	(0.21, 0.47)
Master/PhD degree	-0.85	0.42*	0.11	(0.25, 0.71)
Family size	0.20	1.23*	0.08	(1.07, 1.41)
History of COVID-19(Ref: No COVID-19 infection)
Yes, hospitalized	-0.75	0.47**	0.10	(0.32, 0.70)
Yes, no hospitalization	-0.35	0.71	0.28	(0.32, 1.53)
Income level (Ref: No income)
< 894.94	0.96	2.61*	0.83	(1.40, 4.87)
894.94–1678	0.66	1.94*	0.57	(1.09, 3.44)
1678-2796.68	0.19	1.21	0.39	(0.65, 2.26)
> 2796.68	-0.43	0.65	0.25	(0.31, 1.39)

CE Coefficient Estimate, SD Standard Error, CI Confidence Interval

* P <0.01, ** P <0.001

Education showed a strong negative correlation with membership in class 2, indicating that higher educational attainment reduced the odds of being in class 2 (associate or bachelor’s degree: OR = 0.32, p < 0.001; master/PhD degree: OR = 0.42, p = 0.001). Family size was positively associated with Class 2 membership (OR = 1.23, p = 0.003).

Health-related factors also were related to class membership. Respondents with very good self-reported health were less likely to belong to class 2 (OR = 0.46, p < 0.01). The history of COVID-19 significantly reduced the likelihood of class 2 membership among those hospitalized due to COVID-19 (OR = 0.47, p < 0.001), whereas non-hospitalized cases did not show a significant association (OR = 0.71, p = 0.38).

Insurance coverage was also a significant determinant. Compared to individuals with no insurance, membership in class 2 was less likely among those with social security insurance (OR = 0.37, p = 0.01), health insurance (OR = 0.41, p = 0.03), and other insurance types (OR = 0.24, p < 0.01). Supplementary insurance was not significantly related to class membership (OR = 1.24, p = 0.23).

Sex (OR = 1.47, p = 0.14), head of household status (OR = 1.24, p = 0.44), home ownership (OR = 0.88, p = 0.45), underlying disease (OR = 1.12, p = 0.60), and employment status (OR = 0.73, p = 0.16) were not statistically significant predictors.

Income level positively was related to Class 2 membership for the first two higher income categories (OR = 2.61, p < 0.01; OR = 1.94, p = 0.02), whereas higher categories did not show significant associations.

Overall, these findings suggest that demographic, socio-economic, and health-related factors—including age, marital status, education, family size, health status, insurance coverage, income, and prior COVID-19 hospitalization—significantly affect the probability of belonging to class 2, demonstrating heterogeneity in respondents’ preferences for hospital in the home COVID-19 care services.

Discussion

This study aimed to identify preferences for hospital in the home care for COVID-19 and to characterize distinct patient segments. Employing a latent-class approach, two respondent segments regarding home-based COVID-19 care were identified, which consistently differed in value placed on price, physician specialty, emergency contact availability, and caregiver qualifications. The logit model further clarified the socio-economic, health, and demographic factors associated with membership in each class.

The results demonstrated significant heterogeneity in preferences. While class 1 showed a relatively larger absolute price coefficient (compared to class 2), their substantially higher WTP for emergency access, clinical specialty and caregiver qualification suggests labeling this class as safety- and quality-oriented, yet price-sensitive, whereas Class 2 exhibited lower price sensitivity but comparatively greater preference for physician specialization.

Consistent with prior expectations, increasing the price of one-day home care reduced the probability of choosing an alternative in both classes indicating that both groups were price-averse. Class1 respondents exhibited considerably greater cost sensitivity than Class 2. Nonetheless, the absolute value of the price coefficient was smaller than that of other attributes in both groups, indicating that while cost matters, respondents valued certain service features more strongly, mirroring findings from a Chinese study [30]. Similarly, in Germany, co-payment was the least influential attribute for selecting home and community long-term care [31]. Another study on integrated care preferences for patients with multiple chronic diseases also found cost to be the least important attribute across all classes [37].

In the membership model, having social security, health, or other basic insurance coverage versus none was associated with lower odds of belonging to class 2 (specialist-oriented), indicating that insured individuals were more likely to be in Class1 (safety- and quality-oriented). Although economic theory suggests that insured respondents are less sensitive to price, the larger absolute price coefficient observed in class 1 suggests greater cost sensitivity among these individuals. This apparent paradox may reflect incomplete insurance coverage [38] and the presence of out-of-pocket [39] or informal payments [40], which maintain the relevance of cost in decision-making. Contrary to our findings, another study reported co-payment as one of the most influential attributes for home-based services [29]. The discrepancy may reflect differences in patient payment patterns for hospitalization: in the Netherlands, hospitalization co-payments were zero, whereas in Iran patients may face substantial out-of-pocket and informal payments [40], with COVID-19-related costs accounting for about 9% of direct costs per patient [39] and unofficial payments beyond formal tariffs [40]. Thus, patients currently contribute substantially to health expenditures, so shifting to home care does not markedly alter their financial burden.

Relative to general physicians, class 1 placed a very high value on combined general- and specialist-care, while class 2 valued it positively but more modestly. A sole specialist did not significantly shift utility in class 1 but increased it in class 2, suggesting that for the safety- and quality-oriented segment, “team-based” care signals higher quality than specialist status alone. Although class 2 participants placed relatively less importance on combination care, having a specialist, either alone or with a generalist, remained among the two most important attributes for this group. International evidence shows similar patterns. A Swiss DCE [41] showed that moving to “all doctors” or “all health professionals” increased choice probability, while GP-only decreased it. A Dutch study found that all classes preferred specialists, and pulmonary nurses (vs. general) increased home hospitalization probability [29]. A German study found quality of care to be the most important attribute [31]. An Australian study on older adults’ urgent-care preferences identified specialist-trained medical professionals as a key attribute [42]. The preference for specialist care, either alone or with general physician, may be explained by the nature of COVID-19 as a novel and severe disease, for which specialist expertise is perceived as critical. Consistently, a systematic review reported better outcomes with specialist care in 24 of 49 studies [43]. This tendency may also reflect more favorable attitudes toward specialists; for example, patients preferred telecare by specialists over in-person care by general physician [44], and valued direct access to specialists without referrals [45]. The membership model showed that higher education levels (associate/bachelor’s; master’s/PhD) were linked to lower odds of belonging to class 2, meaning more educated respondents tended to be in Class 1. This aligns with a Canadian study (lower education was related to less specialized care use) [46], supporting the idea that educated respondents favor broader clinical expertise. DCEs consistently show strong preferences for specialist input [30] or multidisciplinary teams, especially for complex needs—for instance, a DCE on specialist cancer-surgery services found participants were willing to travel longer to access centers offering multidisciplinary specialist care [47]. Together, these findings suggest that, while specialist status matters, the combination of team-based care and multidisciplinary expertise is particularly valued by the higher-educated, safety and quality-oriented segment.

Emergency access showed the largest coefficient in Class 1 and was also among the relatively more influential attributes in Class 2. Membership results offer a plausible mechanism: older age groups and those with prior COVID-19 hospitalization were less likely to be in class 2 probably consistent with higher perceived risk and a premium on rapid access. Older respondents, therefore, tended to cluster in class 1, where emergency support is paramount. This finding mirrors prior studies showing that elderly patients often use emergency services more frequently [48]. Additionally, higher education reduced the probability of belonging to class 2, implying that more educated participants recognize COVID-19 risks and value emergency contact more [49]. As a German study shows those with higher education had greater COVID-19 knowledge and were more likely to engage in protective behaviors such as hand hygiene and social distancing [49]. Thus, higher education may be linked to increased risk awareness and emphasis on emergency contact preparedness. Another study highlights the importance of emergency access in home- or community-based care, with participants willing to travel about 3 h to see a specialist surgeon with 24/7 coverage [47]. These findings underscore that timely and continuous access to specialized care is highly valued and should be a key feature in discrete choice experiments assessing home-based health services. In addition, participants with very good or good self-rated health were less likely to belong to class 2. This may reflect healthier individuals’ stronger orientation toward prevention and investment in their health. They may perceive attributes such as having emergency contact as a form of preventive safeguard, thus valuing it more strongly than other considerations. Consistent with this interpretation, a study from Switzerland demonstrated a positive association between better self-rated health and engagement in preventive health behaviors [50].

However, a Chinese study found that increasing nurse competencies reduced the probability of choosing long-term home care [51]. In contrast, our findings showed that both latent classes preferred registered nurses, although the magnitude of this preference was substantially larger in Class 1 than in Class 2. This discrepancy may reflect differences in the perceived clinical relevance of professional nursing care in the context of COVID-19 compared with long-term care settings, where non-registered caregivers often play a more prominent role in providing daily support [52]. In our study context, respondents may have associated registered nurses with higher clinical safety and infection control, particularly given the uncertainty and perceived risk of COVID-19 hospital in the home care. Our logistic regression showed that respondents with a history of COVID-19 infection and hospitalization were less likely to be in class 2, possibly distrustful of non-hospital settings and favoring formal professional oversight—hence clustering in class 1. Another study found that 74% of participants trusted physicians and formal settings for COVID-19 treatment [53]. Additionally, married individuals were more likely to be in class 2, suggesting they may rely on family experience and social support [54] and prioritize physician specialty and emergency contact over caregiver qualifications. Other studies have shown divorced, widowed, or never-married individuals perceive less social support than married individuals [55]. However, this finding should be interpreted with caution due to its borderline statistical significance and requires further investigation. Increasing family size also raised the likelihood of class 2 membership, which may have a mechanism like marital status on home care preferences. In the USA, greater caregiver burden led caregivers, especially socio-economically disadvantaged caregivers, to prefer skilled-nursing facilities over home-care agencies [32].

Before discussing the implications of the latent class results, it is important to assess the representativeness of the study sample. When compared with the most recent Iranian national census (2016, published in 2018) [56] for the urban population of Tehran, the demographic characteristics of the study sample were broadly comparable, although some statistically significant differences were observed. No significant differences were found for sex (52.7% vs. 50.1%, p = 0.107) or home ownership (47.8% vs. 48.7%, p = 0.58). In contrast, the sample included a higher proportion of married individuals (73.0% vs. 63.4%, p < 0.001) and a larger mean household size (3.67 vs. 3.09, p < 0.001). More notable differences were observed in education, with a higher proportion of bachelor’s degree holders (44.2% vs. 27.1%, p < 0.001) and a lower proportion of individuals with 8–12 years of schooling (27.2% vs. 46.5%, p < 0.001), indicating that the sample was relatively more educated than the general urban population. Overall, while the sample captures key demographic features of the target population, the higher educational attainment should be considered when interpreting the findings and may limit external validity.

The findings of this study have direct implications for health system policymakers, insurance organizations, and healthcare planners in Iran and similar settings. Policymakers can use these results to design patient-centered home-based COVID-19 care models that align with population preferences, particularly by prioritizing emergency access, clinical specialization, and caregiver qualifications.

Insurance providers may use the estimated willingness-to-pay values to inform reimbursement strategies and cost-sharing policies for home-based care services. In addition, healthcare planners can use the identified preference heterogeneity to develop differentiated service packages tailored to distinct population segments, such as high-risk individuals who prioritize safety and emergency responsiveness versus cost-sensitive groups.

These findings are particularly relevant in the context of future pandemic preparedness and health system resilience. Given the risk of recurrent infectious disease outbreaks and hospital capacity constraints, understanding public preferences for home-based care is essential for designing acceptable and effective alternatives to hospital admission.

This study has several limitations. First, the sample was limited to residents attending urban health centers in Tehran city, which may limit generalizability to rural populations or other regions. In addition, convenience sampling within health centers may introduce selection bias. Moreover, the sample was relatively more educated than the general urban population, which may affect external validity and preference estimates. Second, although a DCE approach allows for the assessment of stated preferences, it may not fully reflect actual behavior in real-world settings. Third, only four attributes were included in the final experiment, which may not capture all relevant factors influencing home-based care preferences. The DCE also lacked an opt-out option, forcing respondents to choose between alternatives when in reality they might prefer neither. Future studies should consider including additional attributes such as digital health support, and caregiver availability time. Finally, preferences may evolve over time, particularly in post-pandemic contexts, which was not captured in this cross-sectional design. Despite these limitations, the study offers valuable insights into public priorities for hospital in the home care during COVID-19 and can inform policymakers in creating patient-centered strategies.

Conclusion

This study identified two distinct preference segments for COVID-19 hospital in the home care using a latent class and membership model. Both groups were price-sensitive, but financial factors had weaker association than safety, clinical specialty and caregiver qualification attributes.

Class 1

Older, more educated, covered by social security/health/other basic insurance, better health status, and previously hospitalized COVID-19 individuals. They placed relatively higher emphasis on emergency access, team-based physician care, qualified nurses, and finally low price reflecting heightened risk awareness and a focus on safety, and clinical quality (specialty and qualification).

Class 2

Tended to include more married individuals (though the marital status finding was at the threshold of statistical significance), those with larger families and had lower income levels. They showed greater cost tolerance and compared to the first group placed less emphasis on all attributes. The most important attribute for this group was specialty of physician, indicating a specialty-oriented view.

Overall, affordability matters, but choices are more strongly driven by perceived emergency responsiveness, clinical specialty and qualification. These heterogeneous preferences suggest that policymakers, insurers, and health planners may consider designing flexible, patient-centered home-care models—for example, safety-oriented packages with emergency access for Class 1 and specialist-oriented packages for Class 2. The results provide actionable evidence for tailored policies in future pandemic preparedness.

Abbreviations

WHO

World Health Organization

DCE

Discrete Choice Experiment

MWTP

Marginal Willingness-to-Pay

WTP

Willingness to Pay

Authors’ contributions

VRD, and MEK participated in the conceptualization of the study. MEK collected the data. VRD and DS led the study design and analysis with contributions from LT, and YH. VRD wrote the first draft. All authors provided feedback on the first draft and agreed on the final draft. All authors take full responsibility for the content of the article.

Funding

The study was conducted with the financial support of Hamadan University of Medical Sciences (Grant No. 140105183546).

Data availability

The dataset is accessible to the corresponding author (VRD) and another researcher (MEK). It can be provided by the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

The study protocol was approved by the Ethics Committee of Hamadan University of Medical Sciences (ethical code: IR.UMSHA.REC.1401.295) and conducted in accordance with the ethical standards of the Declaration of Helsinki. Written informed consent was obtained from all participants.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.