http://orcid.org/0000-0001-6712-5535
Figures
Abstract
Health-related quality of life (HRQOL) instruments are sometimes used without explicit understanding of which HRQOL domains are important to a given population. In this study, we sought to elicit an importance hierarchy among 27 HRQOL domains (derived from the general population) via a best-worst scaling survey of the population in Singapore, and to determine whether these domains were consistently valued across gender, age, ethnicity, and presence of chronic illnesses. We conducted a community-based study that sampled participants with quotas for gender, ethnicity, age, presence of chronic illness, and interview language. For the best-worst scaling exercise, we constructed comparison sets according to a balanced incomplete block design resulting in 13 sets of questions, each with nine choice tasks. Each task involved three HRQOL domains from which participants identified the most and least important domain. We performed a standard analysis of best-worst object scaling design (Case 1) using simple summary statistics; 603 residents participated in the survey. The three most important domains of health were: “the ability to take care of self without help from others” (best-worst score (BWS): 636), “healing and resistance to illness” (BWS: 461), and “having good relationships with family, friends, and others” (BWS: 373). The 10 top-ranked domains included physical, mental, and social health. The three least important domains were: “having a satisfying sex life” (BWS: -803), “having normal physical appearance” (BWS: -461), and “interacting with others (talking, shared activities, etc.)” (BWS: -444). Generally, top-ranked domains were consistently valued across gender, age, ethnicity, and presence of chronic illness. We conclude that the 10 top-ranked domains reflect physical, mental, and social dimensions of well-being suggesting that the sampled population’s views on health are consistent with the World Health Organization’s definition of health, “a state of complete physical, mental and social well-being and not merely the absence of disease or infirmity”.
Citation: Uy EJB, Bautista DC, Xin X, Cheung YB, Thio S-T, Thumboo J (2018) Using best-worst scaling choice experiments to elicit the most important domains of health for health-related quality of life in Singapore. PLoS ONE 13(2): e0189687. https://doi.org/10.1371/journal.pone.0189687
Editor: Nan Luo, National University of Singapore, SINGAPORE
Received: April 5, 2017; Accepted: November 30, 2017; Published: February 8, 2018
Copyright: © 2018 Uy et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: All relevant data are within the paper and its Supporting Information files.
Funding: This research is supported by the Singapore Ministry of Health’s National Medical Research Council under its Health Services Research Grant (HSRG/0034/2013), URL of funder's website: www.nmrc.gov.sg, HSR Grant Number: HSRG/0034/2013. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
Introduction
Many instruments have been developed, validated, and used to measure health-related quality of life (HRQOL). Generic health status instruments that can be used among people without illness and among different patient groups are widely used; such instruments allow health state comparisons and valuations to guide healthcare decisions at the individual and population level. These include the instruments from the Patient-Reported Outcomes Measurement Information System (PROMIS), the World Health Organization Quality of Life (WHOQOL) group, Short Form 36 (SF-36) of the Medical Outcomes Study, and the EuroQOL five-dimension questionnaire(EQ-5D), to name a few. Despite the popularity of these instruments, there are notable limitations: First, most of these instruments have been developed and used in Western populations and, only later, adapted for use in other populations. The WHOQOL, having been developed in multiple countries worldwide from its inception, is a notable exception [1–4]. Second, despite painstaking efforts to cross-culturally adapt these instruments prior to their use in other countries, there is mounting evidence that these HRQOL instruments do not adequately account for the differences in culture, health experiences, and priorities in other populations, most especially in Asian countries [5–12]. In addition, these instruments were often developed using a “top-down” approach, with the published literature and the experts in the field guiding how HRQOL is conceptualized and which domains are developed into instruments. This is also evident among HRQOL instruments developed in Asia [13–15].The process of developing HRQOL instruments is at times carried out without formal inquiry into which domains are of importance to the patients and the members of a given population, the “end-users” who are envisioned to use the developed HRQOL instruments. Patient input is sometimes lacking at the point of framework development and identification of domains to be developed into an instrument, and is often limited to reviewing adequacy of domain coverage and the wording of items [1–4].
To address these gaps, we sought to determine the domains of health that are important to Singaporeans by using a “ground-up” approach. With a population of 3.87 million citizens and permanent residents, Singapore is a microcosm of Asia—a multi-cultural, multi-ethnic country with individuals of Chinese, Malay, and Indian ethnicities communing with each other and using English as the lingua franca [16]. Developing an HRQOL instrument in this setting provides a unique opportunity to explore cultural variation in conceptualizations of health and develop a unified instrument to measure HRQOL in this population.
In a previous qualitative study based on focus group discussions and in-depth interviews, we developed a framework to capture the domains of health which are important to the multi-ethnic population in Singapore [17]. The framework identified 27 HRQOL domains. In the current study we sought to determine which of these 27 health domains are perceived to be the most important to subjective wellbeing, from the point of view of the population in Singapore, and to determine whether these domains are consistently valued across gender, age, ethnicity, and presence of chronic illnesses.
Methods
This study underwent ethical review and was approved by the Singhealth Centralised Institutional Review Board.
Study design
The best-worst scaling survey was conducted using a cross-sectional study design. To elicit the “importance ranks” among 27 HRQOL domains for the Singapore population, we administered a best-worst scaling (BWS) survey to sampled, community-dwelling participants. Singapore citizens or permanent residents, 21 years or older, of Chinese, Malay or Indian ethnicity, who spoke either English or Chinese (Mandarin) were eligible.
Creating linguistically-simple domain definitions for ranking
We previously identified 27 health domains that spanned physical, mental, and social health using focus groups and in-depth interviews with Singapore citizens and permanent residents [17]. To standardize interpretation and facilitate comprehension of the health domains during ranking, we used the transcripts from the previous study to develop linguistically-simple English definitions for the 27 domains. To test the comprehensibility of the English domain definitions, we conducted cognitive interviews with convenience-sampled participants of varying age, level of education, gender, and ethnicity from the specialist outpatient clinics of the Singapore General Hospital. The domain definitions were revised iteratively based on inputs from the cognitive interviews. Each iteration was based on the input from three to four cognitive interview participants. Thereafter, we developed a Chinese version of the domain definitions through a forward and back translation process [18]. As in the English version, we performed cognitive interviews with native Chinese speakers and revised the wording of the definitions accordingly. Participants ranked the domains based solely on these definitions; no domain labels or other concept clarifications were provided either verbally or in writing.
Best-worst scaling (BWS)
We employed “object case” or “Case 1” BWS to determine the importance hierarchy of the 27 health domains. In BWS, the best and worst choices represent extremes of an underlying latent or subjective dimension of interest. In a Case 1 BWS task, participants are asked to identify which of a fixed set of “objects” they find the best (“most attractive”, “most useful”, or “most important”) and the worst (“least attractive”, “least useful”, or “least important”) [19, 20]. The mathematical models and proofs for Case 1 BWS are provided in Marley and Louviere [21].
In this study, the BWS exercise had 27 “objects”. Each health domain, represented by a linguistically-simple domain definition, was one “object”.
Construction of comparison sets.
Our goal was to elicit the three highest valued domains among 27 HRQOL domains for the Singapore population as a whole, and within subgroups of gender, age, ethnicity, and presence of chronic illness. We used the algorithm of Khare and Federer [22] to construct comparison sets based on a balanced incomplete block design (BIBD) for 27 objects generating 13 versions (replicates), each containing nine choice tasks (blocks) of size three (i.e., three health domains per choice task) (S1 Table). In this particular BIBD, each health domain appears 13 times and every pair of health domains appears once (concurrence of one). The position of the item in a choice set (i.e., 1st, 2nd, or 3rd from the top of the page) is rotated across the 13 versions with each item appearing 4 to 5 times in each position. Each participant completed two versions, randomly selected, without replacement from the universe of 13 versions. Hence, each participant completed 18 choice tasks in total. In each choice task, participants were asked to identify the most important and the least important health domain. The most important health domain was coded as “1” and the least important as “-1”. We piloted and refined the BWS survey instrument with a convenience sample of participants from the community (n = 20) and specialist outpatient clinics (n = 45).
There are various challenges to obtaining rankings and ratings for a large set of conceptually complex items. In this situation, BWS has various advantages over standard methods of ranking or rating a full set of items. Considerable cognitive burden is involved in ranking a large set of concepts. This makes simply ranking a full list of items unlikely to yield useful information. By presenting only a manageable number of items at any one time (at most 4 or 5 items), BWS provides item ranks across a full set of items. By rating each item, BWS scaling provides information on the magnitude of the importance of each of the items, but also provides information on the importance of the items relative to each other (i.e., it minimizes situations where all items receive equal ratings suggesting equal magnitude of importance even when this is not truly the case) [23,24]. The BWS approach also minimizes limitations associated with numeric rating scales such as the influence of culturally-dependent differences in the interpretation of rating scales and/or personal response styles [18–20], or the risk of acquiescent response bias, which is the tendency to agree rather than disagree regardless of item content. Across a sampled population, the design of choice sets also sets the number of times that each item appears alongside each of the other items and rotates the position of an item so that it appears a set number of times in each position. These limit the influence of the way in which items are presented (i.e., groupings in the set and item position) on the resulting item rank. Despite these advantages, BWS also has limitations: 1) although the rotation of items and item positions minimizes the influence of item presentation on the resulting item ranks, ranking a large set of items using an exhaustive BWS design could mean presenting an unfeasibly large number of choice sets to respondents, 2) respondents are only able to rank subsets of items at any one time without considering the entire set of items to be ranked–whether or not respondents would interpret and rank full sets of items similarly is not clear.
Sampling of participants for community-based survey.
We commissioned a survey company, Nielsen, to recruit participants, administer the survey, and collect survey data.
Participant selection was made according to a multi-stage, sampling plan. The survey company used a proprietary sampling frame that grouped households into geographical clusters, which served as primary sampling units. The sampling frame, comprised of 825,000 households, represented 80% of the non-condominium/non-apartment residential households in Singapore (88% of Singapore residential households are non-condominium/non-apartment dwelling types). Due to access difficulties into condominiums and apartments (12% of Singapore residential households), these gated communities were excluded from the survey. The profiles of the individuals residing in the excluded dwelling types were similar to those who were residing in landed properties which were included in the survey sampling frame [17].
The first stage of sampling involved a simple random sample to select 39 primary sampling units. The second stage of sampling involved a systematic sample to select households; interviewers went door-to-door using fixed route rules and skip patterns to identify candidate households for inclusion. Within each identified household, a single, eligible participant was selected for interview (third stage sampling).
To ensure representation of the demographic segments of interest, we implemented quotas based on gender, age, ethnicity, presence of chronic conditions, and language of interview. The list of chronic conditions used for specifying quotas was based on the Singapore Burden of Disease Study (2010) [25].
Data collection via community-based survey.
Interviewers underwent a preparatory briefing to harmonize the understanding of concepts (e.g. study objectives and definition of terms in the survey instrument) and standardized criteria and/or methods pertinent to participant selection, administration of the best-worst scaling exercise using standardized show cards, and entry of data into a handheld data capture system. Interviewers went door-to-door and collected all data directly from the survey participant. Interviews were conducted from March 21 to May 11, 2015, 10:00 AM to 10:00 PM, daily; with up to three visit attempts (first attempt with two follow-up visits, at least one follow-up visit conducted either in the evening or the weekend) for each identified household.
In the BWS exercise, prior to being shown each choice set, participants were asked the question “Which ONE is the most important for you to be happy and satisfied with life?”. After indicating a choice, participants were then asked “Which ONE is the least important?”. To ensure that participants understood the mechanics of the ranking exercise, the interviewers administered sample choice sets before proceeding to the actual BWS survey instrument. The instructions for the BWS exercise are simple and remain the same throughout the 18 choice sets. As such, the survey did not require an interview guide. However, interviewers were equipped with standardized show cards which showed the two stem questions and one choice set (containing 3 choices) printed in large font, on A4-sized paper (S1 Fig). Interviewers were instructed to read the questions and choices exactly as shown in the show card and were not allowed to reword, or expound on the stem questions and the items in the choice sets.
Fifteen interviewers carried out the original survey. With the exception of one interviewer who only recruited one participant, each interviewer recruited 45 to 48 participants. We tested for interviewer effects by comparing the proportion of participants who were unable to rank (UTR) a number of choice sets: 0, 1, 2, …, up to 18 (the total number of choice sets). We called this the interviewer-specific UTR distribution. We formally tested (using the Kruskal-Wallis non-parametric test) for comparability of the UTR distributions across 14 interviewers and identified one interviewer with an extreme distribution, for whom 100% of participants were unable to rate all 18 choice sets. After identifying a breach in the interview protocol, we deemed the data to be invalid and asked for a full replacement sample. The recruitment of replacement participants was approved by the institutional review board and was carried out from May 9 to 19, 2016 by two interviewers who were involved in the original survey. Each replacement respondent was matched, based on age, gender, and ethnicity, to an original participant.
Statistical analysis.
We followed the standard analysis of best-worst object scaling designs (Case 1) which is based on the best-worst score (B-W) score, as presented by Louviere and colleagues [26]. The B-W score of a health domain is obtained by subtracting the number of times the domain was chosen as least important from the number of times it was chosen as most important across participants. Positive values of the B-W score indicate that a health domain was chosen more frequently as most important, and negative values indicate that a health domain was chosen more frequently as least important. We calculated the average B-W score by dividing the B-W score by the number of participants as all 27 health domains appear with a frequency of two in each participant interview.
Another descriptive measure of importance is the ratio score derived by taking the square root of the total best score divided by the total worst score, i.e., . The resulting coefficient indicates the choice probability relative to the most important domain [21, 27–30]. To facilitate interpretation and comparison across health domains, the square root of (B/W) for all health domains was then scaled by a factor equal to the maximum square root of (B/W) so that the most important health domain was indexed at 100. For simplicity, these scaled quantities are referred here onwards as “index values”. Higher index values represent greater importance ranks.
To further ascertain if choices were consistent across participants, we calculated the standard deviation (SD) of B-W scores and the coefficient of variation (CV) (ratio of the SD to the average B-W score). Higher absolute values for the coefficient of variation represent greater choice heterogeneity.
To compare the top-ranked health domains across subgroups of gender, age, ethnicity, and presence of chronic illnesses, we used a combination of the index value and the CV. Higher index values combined with smaller CV values imply that there is a high concurrence within a group in the choice of a particular health domain as the most important; smaller index values combined with smaller CV values imply a high concurrence within the group in the choice of a particular domain as the least important.
As a sensitivity analysis, we estimated an exploded logit model [31] to compare the importance ranks using this method with those obtained using the standard analysis. We utilized the B-W score of the health domain (at the participant level) as the dependent variable and a set of 26 dummy variables as independent variables, one for each health domain with the domain “self-care” (#14) as reference class. We used the SAS PHREG procedure to output the regression coefficients (betas) and odds ratio. The beta estimate and odds ratio associated with “self-care” were 0.0 and 1.0 respectively. Health domains with smaller betas or smaller odds ratio are associated with a higher standing in the importance hierarchy. To verify agreement (i.e., consistency) between the two ranking approaches, we computed Pearson’s correlation coefficient between the domains’ total B-W score and the beta estimates. A correlation coefficient with absolute value equal to or greater than 0.90 was deemed acceptable to confirm agreement.
Results
Creating linguistically-simple definitions for the domain-ranking exercise
The demographic characteristics of the English and Chinese language cognitive interview participants are summarized in Table 1. To test comprehensibility of the English definitions, we conducted cognitive interviews with convenience-sampled English-speaking participants (n = 29) from the specialist outpatient clinics of the Singapore General Hospital. This resulted in revisions to the wording of the English definitions, which were finalized after five iterations. Thereafter, we developed a Chinese version of the domain definitions through a forward and back translation process [18]. As in the English version, we performed cognitive interviews with native Chinese speakers (n = 12) and revised the wording of the definitions accordingly. We finalized the wording of the Chinese domain definitions after four iterations. The set of 27 health domains and corresponding definitions, as presented to participants in the English-language survey, is given in Table 2.
Participant eligibility and response rate
We pre-selected and approached 7591 households; the residents of 4013 (53%) households did not answer the door despite repeated callbacks; 2896 (38%) households had residents who fit the survey inclusion criteria. Of these 2896 households, 92% had residents who were willing to participate in the survey. However, only 603 (21%) households were actually able to participate in the survey due to full quotas or other logistic challenges (e.g. resident was unavailable during the interview appointment). Interviews lasted a median of 29 minutes and ranged from 8 to 174 minutes.
At first glance, the wide range in the duration of interviews appears alarming. However, data from the cognitive interviews and pilots conducted at the start of this study showed that highly literate participants could complete the repetitive BWS exercise within a short period of time. Also, interviews were carried out in the respondents’ homes. As such, many respondents may have had interruptions in the course of the interview (e.g. children requiring attention). However, the electronic data capture system used by the survey company does not allow the timer to be paused during such interruptions, which could have led to artificially prolonged interview duration. In addition, the survey recruited participants of all ages; elderly patients (in both English and Chinese surveys) tended to have longer interview durations.
Socio-demographic characteristics of sampled population.
Six hundred and three (n = 603) residents participated in the BWS survey. Table 3 summarizes the socio-demographic attributes of the survey participants included in the analysis (555 participants from the original sample, 48 participants recruited as replacements). The mean age of the respondents was 45.4 years (SD: 16.3, range: 21 to 88). Many participants reported having “personally had any significant experience with poor health or illness” (17.1%), having a “close friend or family member who had poor health or illness” (39.1%), or having “provided care to a close friend or family who had a chronic illness” (17.7%). Of those who reported having provided care to a close friend or family member (n = 107), 38.3% reported that they were the primary caregiver.
Best-worst scaling survey result
Results (Table 4) revealed that self-care or “being able to take care of self-care needs without the help of others” was consistently considered the most important domain to one’s health-related quality of life (Domain #14, Total B-W score = 636, Index = 100, CV = 1.0). The following domains were likewise consistently considered important based on their relatively high index values and low CVs: healing and resistance to illness or “not falling sick easily and getting well quickly when you are sick” (Domain #4,Total B-W score = 461, Index = 71.4, CV = 1.6), social relationships or “having good relationships with family, friends and others” (Domain #26, Total B-W score = 373, Index = 64.9, CV = 2.0), physical fitness and mobility or “being able to carry out physical activities and move around without difficulty” (Domain #3,Total B-W score = 350, Index = 63.2, CV = 2.0), and personal freedom or “being independent and in control of your life” (Domain #18,Total B-W score = 194, Index = 50.3, CV = 4.0). The domain on sex or “having a satisfying sex life” was consistently considered least important (Domain #9, Total B-W score = -803, Index = 9.3, CV = 0.7). The following domains were also perceived as least important as evidenced by negative B-W scores, relatively low index scores and relatively low CVs, suggesting that these domains had a small chance of being chosen as most important: “having normal physical appearance (acceptable weight, looking healthy)” (Domain #1, Total B-W score = -461, Index = 23.1, CV = 1.7), social contact or “interacting with others (e.g. talking, shared activities, etc.)” (Domain #25, Total B-W score = -444, Index = 22.9, CV = 1.7), speech or “being able to speak clearly so that others will understand (e.g. no slurring of speech)”(Domain #13, Total B-W score = -269, Index = 26.9, CV = 2.7), and sense of growth or “being able to improve in knowledge, skills, and emotional maturity” (Domain #19, Total B-W score = -232, Index = 29.6, CV = 3.3).
In general, the health domains found at the extremes of the importance continuum at the population level (i.e., the most and least important) were reproduced at the subgroup levels (see Tables 5 and 6). This was most evident in the domains of “self-care” and “having a satisfying sex life” which were ranked respectively as most important (i.e., highest index and very low CV values) and least important (i.e., lowest index and very low CV values) across all subgroups of gender, ethnicity, age groups, presence of chronic illness, and language of interview. The following domains were consistently retained among the most important across the subgroups: Domain #4 “not falling sick easily and getting well quickly when you are sick”, Domain #26 “having good relationships with family, friends and others”, and Domain #3 “being able to carry out physical activities and move around without any difficulty”. The following domains were consistently retained among the least important across subgroups: Domain #1 “having a normal physical appearance”, Domain #25 “interacting with others (e.g. talking, shared activities, etc.)”, and Domain #13 “being able to speak clearly so that others will understand (e.g. no slurring of speech)”.
The sensitivity analysis showed an agreement between the ranking approaches based on simple summary statistics and the exploded logit model in the domains that were perceived as most and least important (Table 3, “Est” and “Odds ratio” columns). The most important domains, those with the highest total B-W scores (or highest index values) were associated with lowest beta estimates or odds ratios. We found a strong negative correlation between the total B-W scores and the beta estimates (r = -0.91).
The proportion of participants who were unable to rank more than three choice tasks (out of 18) was 6.1%. The most prominent reasons were that the domains in the choice task were either all equally important (4.3%) or not comparable (4.1%). Given that the five top-ranked domains were consistent across subgroups and the clear (dominant) preferences at the aggregate level (as seen by the separation of the BW scores between domains in the top tertile compared with the middle or lowest tertile), we believe that the small proportion of participants who were unable to rank more than three choice tasks did not affect the overall rankings.
Discussion
We sought to determine which of the different aspects of health Singaporeans perceived to be the most important to their quality of life using domains from an HRQOL framework developed from the ground-up. We found that the participants consistently ranked the following domains as the most important aspects of health: self-care, healing and resistance to illness, social relationships, physical fitness and mobility, and emotions, and that the highest and lowest ranked domains were consistent for age, gender, ethnicity, presence of chronic illness, and language of interview. To the best of our knowledge, this is the first such study in Asia ranking the importance of the different aspects of health in the general population using best-worst scaling, which minimizes variations in the way participants interpret the meaning of scales and response options, thereby ensuring accurate measurements of item ranks.
Globally, studies that rank different aspects of health are few and far between, and our data add to the literature on this topic. Our results align with other importance-hierarchy studies done worldwide by the WHO, and in Japan and Great Britain; this suggests that the highest ranked domains of health may be similar in various socio-cultural contexts. An analysis of the importance-hierarchy of life aspects was carried out using the data collected during the field trial of the WHOQOL group in the 1990s. In the field trial, 4804 respondents across 15 centers worldwide (seven developed countries, eight developing countries) were asked to indicate how important each of a list of life aspects were and how much it affected their QOL using a 5-point Likert scale with descriptors that ranged from “not important” to “extremely important”. The most important aspects of life in the overall analysis, and for the centers in developed countries (United Kingdom, Spain, Australia, France, United States, Netherlands, and Japan), were daily living activities, having energy, overall health, happiness and enjoyment of life, and to move around [33]. The WHOQOL-BREF importance items on daily living activities, happiness and enjoyment of life, and to move around corresponded with our own items of self-care, emotions, and physical fitness and mobility respectively.
Using methods similar to the previous WHOQOL study, the WHOQOL measures of quality of life for use with older adults (WHOQOL-OLD) pilot study, which recruited 7401 adults, 60 years and older across 22 countries, provided further information on the importance-hierarchy of the WHOQOL facets. Among the countries studied were three urban populations in Asia: Tokyo, Japan; Guangzhou, China; and Hong Kong [34]. The most important aspects of life in these urban populations were: overall health, freedom from pain, energy, restful sleep, activities of daily living, availability of medicines and treatment, sensory abilities, and autonomy [35]. A survey conducted in Japan in 1997 (n = 1096) asked respondents to 1) rank (from 1 to 10) each of the ten pre-selected quality of life domains according to the order of importance in their lives and then 2) provide importance scores to each of the items on a scale of a hundred. This study showed that respondents valued personal health and relationship with family as the most important [36]. A survey that recruited 2033 individuals from representative households in Great Britain in 1992 asked participants to spontaneously mention five “most important things in their current life (both good and bad)”; participants were then asked to rank order the items according to importance. This study showed that respondents were most likely to freely mention relationships with family or relatives, their own health, the health of a close person, and finances/standard of living/housing (in that order) [37].
Although previous surveys have shown that personal health is among the most important aspects perceived to contribute to quality of life across many countries, including Singapore [38, 39], they do not provide insight on which specific aspects of perceived health were thought to be most important. In surveys conducted across almost two decades, Singaporeans have consistently chosen health as the one thing they wanted most [39, 40]. Identifying the specific aspects of health that Singaporeans value the most, and targeting improvement in those aspects of health, could potentially lead to greater improvements in HRQOL. In addition, surveys have shown that personal health and family, along with having a comfortable home and a job, were the aspects of life most valued by Singaporeans [39]. Three of the five highest ranked domains in our study were items relating to physical health; our item for social relationships, defined as “having good relationships with family, friends and others”, includes family relationships.
Sex, which was defined as “having a satisfying sex life” in our study, was consistently selected as the least important of the 27 domains ranked in this study. This is consistent with the results of the WHOQOL-BREF study which showed that “sexual life” was consistently ranked the lowest, be it in the overall analysis, the analysis using data from developed countries, or the analysis using data from developing countries (Zimbabwe, Thailand, India (two centers), Panama, Israel, Croatia, Russia) [33]. Interestingly, respondents from six of the seven developed countries in the WHOQOL-BREF ranked “sexual life” as the least important of 25 items of health [33]. Given that our ranking survey was interviewer administered, it is possible that respondents, members of a conservative, Asian society, may have been wary of selecting “having a satisfying sex life” as the most important domain. However, the majority of respondents in the WHOQOL-BREF completed the questionnaire unassisted, making socially-desirable responses less likely. Despite this, participants in the WHOQOL-BREF still ranked “sexual life” as the item with the lowest importance to their life. This would seem to support that sexual life may truly be of low priority for the surveyed respondents.
The majority of studies determined the importance-hierarchy of quality of life domains by using simple rank ordering (i.e., of X items, rank each one in order of importance from 1 to X) or by assigning scores to each of the proffered domains, either through a categorical (e.g. Likert scale) or continuous scale. There is considerable cognitive burden with the former which thus limits the number of items that a given participant can consider at any one time. The latter, on the other hand, is subject to the influence of culturally-dependent differences in the interpretation of rating scales and/or personal response styles [18–20] or the risk of acquiescent response bias. By inquiring on the extremes of a latent construct, BWS overcomes these stated limitations and minimizes the chance that a number of objects are perceived as all “extremely important” [21,23].
We recognize several limitations of this study. First, we used object-case (Case 1) best-worst scaling to establish which aspects of health are most important to Singaporeans. The robust method allowed us to present a large number of items, 27 domains, without overwhelming respondents with a high cognitive burden. By presenting groups of only three domains at any one time, we increased the likelihood that respondents were able to provide us meaningful, thought out responses. However, the method also meant that respondents were not able to consider all 27 domains together and make a conscious decision on how each item is ranked relative to all the other items. Second, the BIBD design we implemented allowed us to establish item ranks across the entire sampled population but not on a “per individual” level.
Flynn has spoken of numerous potential applications of object-case BWS to establish relative importance in healthcare [23]. However, to our knowledge, only one other study used an object-case BWS exercise to establish both the relative importance and the magnitude of the importance across different aspects of health. This study used object-case BWS to establish the importance of different aspects of activities of daily living in the Disability Assessment for Dementia (DAD) scale as reported by the caregivers of patients with Alzheimer’s Disease (AD) (n = 403 (USA), n = 400 (Germany). The study found that the caregivers of patients with AD valued preserved basic activities of daily living (e.g., using the toilet without accidents) over instrumental activities of daily living (e.g., handling money) [24].
Conclusion
We found that respondents in our survey consistently ranked self-care, healing and resistance to illness, social relationships, physical fitness and mobility, and emotions as the most important aspects of health. Our results are similar to those from the few previous studies in this area, suggesting that the highest ranked domains of health may be similar in various socio-cultural contexts.
Supporting information
S1 Table. Design of best-worst scaling choice sets (13 versions).
https://doi.org/10.1371/journal.pone.0189687.s001
(DOCX)
Acknowledgments
We thank Prof Eric Finkelstein and Assistant Prof Semra Ozdemir Van Dyk from the Duke-NUS Medical School, Singapore, for sharing their experience on the design and implementation of best-worst scaling and discrete choice experiments.
We thank Ms Clarice Chan Zi Ying, Mr Yang Hui, and Mr Tan Wee Boon, research coordinators from the Singapore General Hospital, for assisting with the cognitive interviews which informed the development of the linguistically-simple English definitions of the 27 domains which were ranked in this study.
We thank Ms Serene AK Ong, Medical Writer, Duke-NUS Medical School, for her help in manuscript preparation.
References
- 1. Cella D, Yount S, Rothrock N, Gershon R, Cook K, Reeve B, et al. The Patient-Reported Outcomes Measurement Information System (PROMIS). Med Care. 2007;45(5 Suppl 1):S3–11.
- 2. Group W. Study Protocol for the World Health Organization Project to Develop a Quality of Life Assessment Instrument (WHOQOL). Qual Life Res. 1993;2(2):153–159. pmid:8518769
- 3. Ware Jr JE, Sherbourne CD. The MOS 36-item short-form health survey (SF-36): I. Conceptual framework and item selection. Med Care. 1992;30(6): 473–483. pmid:1593914
- 4. EuroQol Group. EuroQol—a new facility for the measurement of health-related quality of life. Health Policy Amst Neth. 1990 Dec;16(3):199–208.
- 5. Li L, Wang HM, Shen Y. Chinese SF-36 Health Survey: translation, cultural adaptation, validation, and normalisation. J Epidemiol Community Health. 2003;57(4):259–263. pmid:12646540
- 6. Thumboo J, Fong K-Y, Machin D, Chan S-P, Leong K-H, Feng P-H, et al. A community-based study of scaling assumptions and construct validity of the English (UK) and Chinese (HK) SF-36 in Singapore. Qual Life Res. 2001;10(2):175–188. pmid:11642688
- 7. Tseng H-M, Jui-fen RL, Gandek B. Cultural issues in using the SF-36 Health Survey in Asia: results from Taiwan. Health Qual Life Outcomes. 2003;1(1):72.
- 8. Fukuhara S, Ware JE, Kosinski M, Wada S, Gandek B. Psychometric and clinical tests of validity of the Japanese SF-36 Health Survey. J Clin Epidemiol. 1998;51(11):1045–1053. pmid:9817122
- 9. Gothwal VK, Srinivas M, Rao GN. A new look at the WHOQOL as health-related quality of life instrument among visually impaired people using Rasch analysis. Qual Life Res. 2013;22(4):839–51. pmid:22648162
- 10. Chung W-S, Lan Y-L, Yang M-C. Psychometric testing of the short version of the world health organization quality of life (WHOQOL-BREF) questionnaire among pulmonary tuberculosis patients in Taiwan. BMC Public Health. 2012;12(1):630.
- 11. Aggarwal AN, Agarwal R, Gupta D. Abbreviated World Health Organization Quality of Life questionnaire (WHOQOL-Bref) in north Indian patients with bronchial asthma: an evaluation using Rasch analysis. NPJ Prim Care Respir Med [Internet]. 2014 [cited 2015 Jul 27];24. Available from: http://www.nature.com/articles/npjpcrm20141?WT.ec_id=NPJPCRM-201406
- 12. Zhao F, Yue M, Yang H, Wang T, Wu J, Li S-C. Validation and comparison of EuroQol and Short Form 6D in chronic prostatitis patients. Value Health. 2010;13(5):649–656. pmid:20412540
- 13. Leung K, Liu F, Zhao L, Fang J, Chan K, Lin L. Development and validation of the Chinese Quality of Life Instrument. Health Qual Life Outcomes. 2005;3(1):26.
- 14. Yeh C-H. Adaptation in children with cancer: Research with Roy’s Model. Nurs Sci Q. 2001;14(2):141–148. pmid:11873328
- 15. Kuo C, Yen M, Lin P. Developing an instrument to measure quality of life of patients with hyperhidrosis. J Nurs Res Taiwan Nurses Assoc. 2004;12(1):21–30.
- 16.
Population trends 2014 [Internet]. Department of Statistics Singapore; 2014 [cited 2015 Aug 27]. Available from: http://www.singstat.gov.sg/publications/publications-and-papers/population-and-population-structure/population-trends
- 17.
Thumboo J, Ow MYL, Uy EJB, Xin X, Chan ZYC, Sung SC, Bautista DC, Cheung YB. Developing a comprehensive, culturally sensitive conceptual framework of health domains in Singapore. Submitted.
- 18. Guillemin F, Bombardier C, Beaton D. Cross-cultural adaptation of health-related quality of life measures: Literature review and proposed guidelines. J Clin Epidemiol. 1993;46(12):1417–1432. pmid:8263569
- 19. Louviere J, Lings I, Islam T, Gudergan S, Flynn T. An introduction to the application of (case 1) best–worst scaling in marketing research. Int J Res Mark. 2013;30(3):292–303.
- 20.
Marley AAJ, Flynn TN, Australia V. Best worst scaling: theory and practice. Handb Choice Model Edw Elgar Publ Leeds UK [Internet]. 2012 [cited 2015 Jul 4]. Available from: http://130.220.0.27/Global/business/centres/i4c/docs/papers/wp12-001.pdf
- 21. Marley AAJ, Louviere JJ. Some probabilistic models of best, worst, and best–worst choices. J Math Psychol. 2005;49(6):464–480.
- 22. Khare M, Federer WT. A simple construction procedure for resolvable incomplete block designs for any number of treatments. Biom J. 1981;23(2):121–132.
- 23. Flynn TN. Valuing citizen and patient preferences in health: recent developments in three types of best–worst scaling. Expert Rev Pharmacoecon Outcomes Res. 2010;10(3):259–67. pmid:20545591
- 24. Hauber AB, Mohamed AF, Johnson FR, Cook M, Arrighi HM, Zhang J, et al. Understanding the relative importance of preserving functional abilities in Alzheimer’s disease in the United States and Germany. Qual Life Res. 2014;23(6):1813–1821. pmid:24448684
- 25.
Singapore Burden of Disease Study 2010 | Ministry of Health [Internet]. [cited 2014 Sep 19]. Available from: http://www.moh.gov.sg/content/moh_web/home/Publications/Reports/2014/singapore-burden-of-disease-study-2010.html
- 26. Louviere JJ, Flynn TN. Using best-worst scaling choice experiments to measure public perceptions and preferences for healthcare reform in Australia. Patient Patient-Centered Outcomes Res. 2010;3(4):275–283.
- 27. Auger P, Devinney TM, Louviere JJ. Using best–worst scaling methodology to investigate consumer ethical beliefs across countries. J Bus Ethics. 2006;70(3):299–326.
- 28. Cohen E. Applying best‐worst scaling to wine marketing. Cohen E, editor. Int J Wine Bus Res. 2009;21(1):8–23.
- 29. Flynn TN, Louviere JJ, Peters TJ, Coast J. Best–worst scaling: What it can do for health care research and how to do it. J Health Econ. 2007;26(1):171–189. pmid:16707175
- 30. Lee JA, Soutar G, Louviere J. The best–worst scaling approach: An alternative to Schwartz’s values survey. J Pers Assess. 2008;90(4):335–347. pmid:18584442
- 31. Allison P. Logit models for sets of ranked items. Sociol Methodol. 1994;24:199–228.
- 32.
Department of Statistics. Population trends 2014. Ministry of Trade & Industry, Republic of Singapore 2014.
- 33. Saxena S, Carlson D, Billington R, Orley J. The WHO quality of life assessment instrument (WHOQOL-Bref): The importance of its items for cross-cultural research. Qual Life Res. 2001;10(8):711–721. pmid:11871592
- 34. Power M, Quinn K, Schmidt S, WHOQOL-OLD Group. Development of the WHOQOL-Old Module. Qual Life Res. 2005;14(10):2197–214. pmid:16328900
- 35. Molzahn AE, Kalfoss M, Makaroff KS, Skevington SM. Comparing the importance of different aspects of quality of life to older adults across diverse cultures. Age Ageing. 2011;40(2):192–199. pmid:21186234
- 36. Tarumi K, Imanaka Y, Isshiki Y, Morimoto K. Quality of life domains in the healthy public: A trial investigation using attendants for an annual health checkup. Environ Health Prev Med. 1999;4(1):39–48. pmid:21432170
- 37. Bowling A. What things are important in people’s lives? A survey of the public’s judgements to inform scales of health related quality of life. Soc Sci Med. 1995;41(10):1447–1462. pmid:8560313
- 38.
Inoguchi T. AsiaBarometer Survey Data 2004 and 2006 [Internet]. AsiaBarometer Project. Asia Barometer is a registered trademark of Takashi Inoguchi,President of the University of Niigata Prefecture, Japan, and Director of the AsiaBarometer Project.; [cited 2015 Oct 22]. Available from: http://www.asiabarometer.org/
- 39.
Tambyah SK, Tan SJ, Kau AK. Priorities in life, top worries and views on government spending. In: The wellbeing of Singaporeans: Values, Lifestyles, Satisfaction and Quality of Life. Singapore: World Scientific; 2010. p. 30–54.
- 40.
Kau AK. Understanding Singaporeans: Values, Lifestyles, Aspirations, and Consumption Behaviors. World Scientific; 2004.