Vol.:(0123456789) Quality of Life Research (2024) 33:1107–1119 https://doi.org/10.1007/s11136-023-03590-z Mapping of Family Reported Outcome Measure (FROM‑16) scores to EQ‑5D: algorithm to calculate utility values R. Shah1  · M. S. Salek2  · A. Y. Finlay1  · R. Kay3  · S. J. Nixon4 · K. Otwombe5  · F. M. Ali1  · J. R. Ingram1 Accepted: 17 December 2023 / Published online: 25 February 2024 © The Author(s) 2024 Abstract Objective Although decision scientists and health economists encourage inclusion of family member/informal carer utility in health economic evaluation, there is a lack of suitable utility measures comparable to patient utility measures such those based on the EQ-5D. This study aims to predict EQ-5D-3L utility values from Family Reported Outcome Measure (FROM- 16) scores, to allow the use of FROM-16 data in health economic evaluation when EQ-5D data is not available. Methods Data from 4228 family members/partners of patients recruited to an online cross-sectional study through 58 UK- based patient support groups, three research support platforms and Welsh social services departments were randomly divided five times into two groups, to derive and test a mapping model. Split-half cross-validation was employed, resulting in a total of ten multinomial logistic regression models. The Monte Carlo simulation procedure was used to generate predicted EQ- 5D-3L responses, and utility scores were calculated and compared against observed values. Mean error and mean absolute error were calculated for all ten validation models. The final model algorithm was derived using the entire sample. Results The model was highly predictive, and its repeated fitting using multinomial logistic regression demonstrated a stable model. The mean differences between predicted and observed health utility estimates ranged from 0.005 to 0.029 across the ten modelling exercises, with an average overall difference of 0.015 (a 2.2% overestimate, not of clinical importance). Conclusions The algorithm developed will enable researchers and decision scientists to calculate EQ-5D health utility esti- mates from FROM-16 scores, thus allowing the inclusion of the family impact of disease in health economic evaluation of medical interventions when EQ-5D data is not available. Keywords FROM-16 · EQ-5D · Utility values · Response mapping · Health technology · Health economic evaluation · Family member/partner · Informal carer · Quality of life * R. Shah ShahR45@cardiff.ac.uk M. S. Salek m.s.salek@herts.ac.uk A. Y. Finlay finlayay@cardiff.ac.uk R. Kay richard.kay@rkstatistics.com S. J. Nixon stuart.nixon@mssociety.org.uk K. Otwombe otwombek@phru.co.za F. M. Ali alifm@cardiff.ac.uk J. R. Ingram ingramjr@cardiff.ac.uk 1 Division of Infection and Immunity, School of Medicine, Cardiff University, Cardiff, UK 2 School of Life & Medical Sciences, University of Hertfordshire, Hatfield, UK 3 RK Statistics, Bakewell, UK 4 Multiple Sclerosis Society, Cardiff, UK 5 Statistics and Data Management Centre, Perinatal HIV Research Unit, University of the Witwatersrand, Johannesburg, South Africa http://orcid.org/0000-0001-8158-712X http://orcid.org/0000-0002-4612-5699 http://orcid.org/0000-0003-2143-1646 http://orcid.org/0000-0001-8968-0951 http://orcid.org/0000-0002-7433-4383 http://orcid.org/0000-0002-4184-2023 http://orcid.org/0000-0002-5257-1142 http://crossmark.crossref.org/dialog/?doi=10.1007/s11136-023-03590-z&domain=pdf 1108 Quality of Life Research (2024) 33:1107–1119 Introduction Public healthcare decision-making is increasingly sup- ported by economic analyses of healthcare interventions. Decision scientists and economists compare the costs and outcomes of novel interventions with the best alternative to inform their cost-effectiveness. Currently, such analysis is focused only on patient outcomes [1]. However, new treatments that improve patients’ quality of life (QoL) can also improve the QoL of their family members. Caring for one’s relative may leave family members/partners physi- cally and emotionally drained. This impact is particularly high where there is a significant amount of caring, such as for a chronic neurological condition. Ignoring the poten- tially large impacts on QoL of family members/partners may result in inequitable and inaccurate evaluation of the medical intervention [2]. Although the inclusion of family member burden in economic evaluations is encouraged by many health technology assessment agencies, such as the National Institute for Health and Care Excellence (NICE), it is seldom reported. Some researchers attribute this to uncertainty about decision-makers’ attitudes toward their inclusion, issues over how the burden may be incorporated into economic models, and the availability of suitable util- ity measures for carers/family members [3]. The lack of carer data may be the most plausible explanation as to why this impact is not included in CEAs [1, 4, 5], as fam- ily member/informal carer inclusion in HTA is a recent recommendation with currently no family members/carer data being collected in clinical trials or alongside patient registries. In the UK, NICE uses the Quality Adjusted Life Year (QALY), a composite measure of quality and quantity of life, to quantify the health effect of a medical interven- tion and ultimately inform resource allocation. In order to generate QALYs, health utilities (or HRQoL weights) are needed, and the NICE preferred measure is the European Quality of Life-5 Dimensions three-Level (EQ-5D-3L) [6]. Some authors argue that generic preference-based meas- ures (PBMs) such as EQ-5D may not be adequate to assess carer utility as they were not designed for this purpose [7]. Although the CarerQoL [8] and Carer Experience Scale (CES) [9] have been valued using choice-based methods, these cannot be used to estimate utility weights [7]. Nev- ertheless, generic PBMs have been used successfully to assess family member/informal carer utilities, with EQ-5D being the most common generic instrument to measure Carer utility [10]. Evidence from a recent study comparing five QoL instruments for carers across four conditions has shown that EQ-5D had some validity and may be appropri- ate for use in health technology evaluations [11, 12]. The main advantage of using the EQ-5D to measure family member/informal carer QoL is that it can easily be com- bined with patient QoL, allowing greater comparability across appraisals. Therefore, mapping family-specific QoL measures such as the Family Reported Outcome Measure (FROM-16) to EQ-5D will allow the inclusion of family members and/or informal carers in health economic evalu- ation when EQ-5D data is not available. The FROM-16 measures QoL impact of a patient’s disease on their fam- ily members/partners across all areas of medicine [13]. It is validated and translated into many languages [14–17]: score descriptor bands have been calculated [18]. Map- ping FROM-16 to EQ-5D would enable the calculation of QALYs for family members and/or informal carers allow- ing comparability with patient utilities. Direct mapping uses either the total or subdomain scores to predict Preference-based measure utility values, while response mapping predicts EQ-5D responses for utilities from the responses on other measures. The most common approach used for direct mapping is the Ordinary Least Square (OLS), that has several limitations. First, it suggests that utilities are continuously distributed and therefore, the utility value of 1.0 cannot be achieved [19]. Secondly, in the case of ceiling effects, OLS can produce inconsistent esti- mates of the coefficients of explanatory variables. Although other methods of direct mapping have been explored to overcome these issues [20, 21], these methods can only provide mapping for a single set of utility values relevant to the country of tariff. In contrast, response mapping predicts EQ-5D dimension responses, which can be used to derive utility values using any country-specific tariff [19]. In this study, we use response mapping to predict EQ-5D health utility estimates from FROM-16 responses to allow the use of FROM-16 in health economic evaluation. Methods Study design and participant recruitment The data came from family members/partners of patients with different health conditions recruited through an online cross-sectional study conducted between April and Novem- ber 2021. Participants were recruited through 58 UK-based patient support groups, research support platforms (Health- wise Wales-[HWW] [22], Autism Research Centre-Cam- bridge University database [ARC], Join Dementia Research [JDR]) and the Welsh social services departments. Family members/partners of patients completed the FROM-16 and EQ-5D-3L questionnaires. Ethics approval was given by the Cardiff University School of Medicine Research Ethics committee (SREC reference: 21/19), conforming to the prin- ciples embodied in the Declaration of Helsinki. The study was open to family members/partners (aged ≥ 18 years) of 1109Quality of Life Research (2024) 33:1107–1119 patients with any health condition and any age or gender living in the UK. Measures FROM‑16 The FROM-16 is a generic family QoL questionnaire which measures the impact of any disease, across all medical spe- cialities, on the QoL of adult family members or partners of patients of any age [13]. The FROM-16 comprises 16 items, each with three response options: ‘Not at All’ (scoring 0), ‘A Little’ (scoring 1) and ‘A Lot’ (scoring 2). The lowest pos- sible score of FROM-16 is 0, and the highest 32. The higher the score, the more negative the family member’s QoL. EQ‑5D‑3L The Euroqol EQ-5D is a generic HRQoL questionnaire which measures preferences associated with a particular health state. The EQ-5D consists of 5 dimensions (mobil- ity, self-care, usual activities, pain, and anxiety), each with 3 levels (no problem, some problems, and extreme problems) coded 1 to 3. The EQ-5D-3L descriptive system presents 243 health states that are combined to calculate a single index, where the best health status is "11111", and the worst "33333". For this study, the index was calculated using the set of specific values (Tariffs) of the EQ-5D-3L UK ver- sion [23]. In this tariff, the utility values attached to differ- ent EQ-5D health states range from − 0.594 to 1, where 1 is defined as perfect health, 0 represents death, and negative values denote health states worse than death. Statistical analysis Exploratory analysis The frequencies and percentages of each response cat- egory of the items of both questionnaires were calculated along with the mean and standard deviation (SD) for the continuous variables. The distributions of the EQ-5D-3L index and FROM-16 dimensions were graphically observed through histograms, and normality was checked using Sha- piro–Wilk’s test. Spearman correlations between the EQ- 5D-3L index and the FROM-16 total score were calculated, defining “moderate” correlation as values between 0.30 and 0.49, “strong” between 0.50 and 0.69 and “very strong” for a value > 0.70 [24]. Mapping the FROM‑16 to EQ‑5D responses We used the multinomial logistic regression (mlogit) to explore the association between individual FROM-16 responses (independent variable) and EQ-5D responses for each dimension (dependent variable). As the dependent vari- ables are ordinal in nature, ordinal logistic regression would be the preferred method. However, the ordered logit model relies on an assumption of proportional odds or parallel regression, which means it generates a set of binary response models for the different ordered categories, in which the intercept varies, but the coefficients for the explanatory vari- ables are the same. We first attempted ordinal logistic regres- sion but found that the assumption of proportional odds was violated for all dimensions of EQ-5D-3L (the test for parallelism within SPSS gave significant results for all five EQ-5D dimensions, indicating violation of the proportional odds assumption). The alternative was therefore mlogit, which avoids the parallel regression assumption and pro- vides unbiased parameter estimates. Using all data, a series of mlogit regressions were fitted for each of the five EQ-5D dimensions against the 16 individual items of FROM-16, as well as age and sex, using SPSS version 27. All 16 FROM- 16 items were included for each domain model to capture all correlations induced by each FROM-16 item. Regressions were run with age and sex alone, FROM-16 items alone, as well as age and sex combined with FROM-16 items (Sup- plementary File 1, Table S1) to evaluate the contribution of age and sex, and collectively the FROM-16 items. Model comparisons were undertaken by comparing twice the abso- lute difference in the maximized log-likelihoods with the Chi-square distribution, with degrees of freedom equal to the difference in the number of model terms being evaluated (Supplementary File 1, Table S1). Split‑half cross‑validation This study employed the Split-half method used by Ali et al. [25] for mapping the Dermatology Life Quality Index (DLQI) to the EQ-5D, whereby the dataset was randomly split five times into separate estimation and validation sets using the SPSS version 27 random number generator. The estimation set was used to derive the mapping models, whilst the out-of-sample validation set was utilised for validating the fitted models. The multinomial logistic regression was conducted for each pair of datasets using FROM-16 items, age, and sex as independent variables. The model was tested on each validation dataset to produce three predicted proba- bilities per subject per EQ-5D domain (Y = 1, 2, or 3). Using these predicted probabilities, a Monte Carlo simulation was carried out for each subject resulting in predicted domain responses and consequently health utility estimates. A Monte Carlo method ensures that unbiased expected values are obtained and allows individuals to be identified within the EQ-5D descriptive system and predicted utility scores or tariffs to be calculated using the UK time trade off (TTO) values [23]. The five estimation and validation sets were 1110 Quality of Life Research (2024) 33:1107–1119 then switched, and the process was repeated (split-half cross- validation), resulting in ten models. The average predicted health utility estimate for each validation set was then com- pared with the observed health utility estimate of the same set. Means square error (MSE) and mean absolute error (MAE) were compared and averaged across 10 validation models. The final model algorithm was based on the entire sample of data from 4228 family members/partners [26]. Results Study sample demographic characteristics A total of 4228 family members/partners of patients across 27 medical specialities, mostly from England and Wales, completed the EQ-5D and FROM-16 questionnaires (Table 1, Supplementary File 1, Table S2a). The mean age of family members was 57.7 (SD = 14.2) years, 65% were female. Patients’ mean age was 61.6 (SD = 20.2) years, 54% female. The family members were mostly the patient’s spouse/partner (60%), sons/daughters (22%) and parents (12%) (Table 1). FROM‑16 and EQ‑5D scores The mean FROM-16 total summary score and the EQ-5D-3L utility score were 14.8 (SD = 8.1) and 0.673 (SD = 0.3) (Table 1). Among FROM-16 items, ‘feeling worried’ was the most rated impact and ‘effect on travel’ was the least rated impact, while on the EQ-5D-3L domains, anxiety/ depression was the most rated problem and ‘selfcare’ was the least rated problem (Supplementary File 1, Tables S2b, S2c). There was no evidence of significant multicollinearity between the sixteen FROM-16 items. For example, the cor- relation coefficient between worry, anger, sadness and frus- tration ranged from 0.424 to 0.593, less than the 0.7 thresh- old for multicollinearity (Supplementary File 1, Table S2d). Characterising the distribution and conceptual overlap Figure 1 shows the distribution plots of the EQ-5D-3L util- ity scores and FROM-16 total scores. FROM-16 appears to be normally distributed while EQ-5D-3L appears to be negatively skewed, indicating non-normality. Although the Shapiro–Wilk’s test was significant for FROM-16 and EQ-5D data sets indicating non-normality, for large sam- ple sizes, histograms are more appropriate [27]. The cor- relation between the FROM-16 total summary score and the EQ-5D-3L utility scores was moderate with a Spear- man’s rank correlation coefficient (rs) of 0.45. The EQ-5D anxiety/depression was strongly associated with FROM-16 domains (Emotional domain r = 0.52; Personal and social domains r = 0.50) while the EQ-5D mobility showed weak- est association with FROM-16 Emotional domain (r = 0.132) (Supplementary File 1, Table S3). The relationship between Table 1 Demographics and descriptive statistics rs Spearman correlation coefficient a Brother/Sister, Father/Mother in law, Grandparent, Uncle/Aunt, Grandson/Granddaughter, Brother/Sister in law, Nephew/Niece, Cousin, friend *Significant at 1% Variables n (%) or M (SD) Family members (n = 4228)  Gender   Male 1479 (35.0%)   Female 2749 (65.0%)  Age (years)   Mean (SD) 57.69 (14.2)   Median 60   Range (IQR) 18–95 (20)  Relationship to patient   Spouse/Partner 2532 (59.9%)   Son/daughters 921 (21.7%)   Parent 503 (11.9%)    Othera 272 (6.4%) Patients (n = 4228)  Gender   Male 1928 (45.6%)   Female 2290 (54.2%)   Prefer not to say 4 (0.1%)   Other 6 (0.1%)  Age (years)   Mean (SD) 61.62 (20.2)   Median 66   Range (IQR) 2–100 (26)  Place of residence in UK   England 1779 (42.1%)   Northern Ireland 45 (1.1%)   Scotland 175 (4.1%)   Wales 2229 (52.7%)  FROM-16 total score (n = 4228)   Mean (SD) 14.79 (8.1)   Range 0–32  EQ-5D-3L utility score (n = 4228)   Mean (SD) 0.673 (0.3)   Range − 0.594–1  EQ-VAS (n = 4209)   Mean (SD) 68.44 (21.9)   Range 0–100  FROM-16 & EQ-5D correlation    rs 0.450* 1111Quality of Life Research (2024) 33:1107–1119 Fig. 1 a Distributions of the EQ-5D-3L utility value; b Distributions of FROM-16 Total Summary Score; c Scat- terplot showing the relationship between FROM-16 total sum- mary score and EQ-5D utility value (c) 1. 00 0 1112 Quality of Life Research (2024) 33:1107–1119 FROM-16 summary scores and EQ-5D utility scores is shown in Fig. 1c (rs = 0.45)”. Split‑half cross‑validation and model performance Five times random split of the entire sample (n = 4228) into two parts resulted in five derivation and five validation sets of 2114 family members each. For each of the five EQ-5D domains, an mlogit model was derived and used to predict the probability of each EQ-5D response for each subject in each validation set using Monte Carlo simulation, and subse- quently, the health utility was estimated. The predicted utili- ties for each validation set were compared to the observed utility. In each case, the predicted mean utility value was lower than the actual mean utility value indicating a slight overestimate of poor health (Table 2). Across the ten validation sets, the difference between actual and predicted mean values ranged from 0.005 to 0.029, with an overall mean difference of 0.015. This 2.2% overestimate represents a clinically unimportant effect; the minimal clinically important difference of EQ-5D var- ies from 0.03 to 0.52 [28]. The mean square error (MSE) across all ten validation sets ranged from 0.132 to 0.141 (mean = 0.137), and the mean absolute error (MAE) across all ten validation sets ranged from 0.266 to 0.275 (mean = 0.269). Table 3 reports error across subset range (EQ-5D < 0, 0 ≤ EQ-5D < 0.25, 0.25 ≤ EQ-5D < 0.5, 0.5 ≤ EQ-5D < 0.75,0.75 ≤ EQ-5D ≤ 1), to further understand variation between observed and predicted utilty values [26]. The smallest difference between observed and predicted mean (ME = − 0.007) was found for 0.25 ≤ EQ-5D subset while the largest difference (ME = − 0.079) in mean was for sub- set EQ-5D < 0. This is consistent with the finding that the degree of error is not evenly distributed across the scale of the dependent variable with overall the level of error being far greater at the lower (more severe health state) end [29]. To test the predictive performance of the equations, EQ-5D responses were assigned using a Monte Carlo approach in which random numbers were compared with the probability values estimated by the mlogit models. Using all FROM-16 questions, age and gender as predictors, the overall proportion of predicted responses allocated to the correct level varied across ten models with most (79%) having 90–100% accuracy. Accuracy was less than 70% (Supplementary File 1, Table S4) in only 4% of responses. In general, predicted levels that were ‘off-diagonal’ were equally likely to be lower or higher than the actual level (Supplementary File 1, Fig. S1). To further compare results, we examined cumulative dis- tribution of observed and predicted utility data across ten validation models (Fig. 2). For models 2, 4, 6 and 8 the predicted distribution was closer to observed data than the other models (Supplementary File 1, Fig. S2). The model was shown to be highly predictive and repeated data splits demonstrated its stability (Table 2). The predictive ability of the model at an individual subject level was examined using histograms to display the difference between the predicted utility score and the actual utility score for each simulation for individual subjects (Fig. 3). The results from all ten splits are displayed in Supplementary File 1, Fig. S3. All the plots depict a centrality around ‘0’, indicating the strong predictive collective capability of the mlogit mod- els. On average across ten validation models, 54% of the individual utility values were predicted to lie within 0.05 of the actual values, 59% within 0.1, 73% within 0.2 and 83% within 0.3 of the actual values (Supplementary File 1, Table S5). We also explored the Expected utility method [30]) which uses predicted probabilities of response levels to directly Table 2 Split-half cross validation using multinomial logistic regression: differences between actual and predicted utility value SD Standard deviation, Diff in means Difference in means, MSE Mean square error, MAE Mean absolute error Cross validation set Actual utility Predicted utility Actual versus predicted Mean (SD) Min Max Mean (SD) Min Max Diff in means MSE MAE Set 1 0.667 (0.342) − 0.594 1 0.662 (0.262) − 0.239 1 0.006 0.135 0.267 Set 2 0.669 (0.331) − 0.594 1 0.655 (0.277) − 0.286 1 0.008 0.136 0.269 Set 3 0.673 (0.331) − 0.594 1 0.646 (0.276) − 0.237 1 0.027 0.140 0.275 Set 4 0.672 (0.326) − 0.594 1 0.667 (0.277) − 0.349 1 0.005 0.138 0.267 Set 5 0.680 (0.326) − 0.594 1 0.654 (0.279) − 0.429 1 0.027 0.132 0.266 Set 6 0.679 (0.320) − 0.594 1 0.659 (0.274) − 0.286 1 0.020 0.135 0.268 Set 7 0.677 (0.331) − 0.594 1 0.648 (0.273) − 0.222 1 0.029 0.140 0.274 Set 8 0.672 (0.330) − 0.594 1 0.667 (0.278) − 0.426 1 0.005 0.141 0.270 Set 9 0.674 (0.336) − 0.594 1 0.657 (0.273) − 0.322 1 0.017 0.138 0.269 Set 10 0.666 (0.336) − 0.594 1 0.658 (0.273) − 0.322 1 0.007 0.136 0.268 Average of 10 Sets 0.673 (0.331) − 0.594 1 0.658 (0.274) − 0.312 1 0.015 0.137 0.269 1113Quality of Life Research (2024) 33:1107–1119 Table 3 Comparison of observed and predicted utilities for EQ-5D subset range Mean EQ-5D-3L Obs/Pre < 0 0 to < 0.25 0.25– < 0.5 0.5– < 0.75 ≤ 0.75 Set 1 Obs − 0.159 0.122 0.341 0.662 0.9 Pre − 0.065 0.150 0.341 0.678 0.86 Set 2 Obs − 0.153 0.126 0.333 0.662 0.897 Pre − 0.086 0.151 0.338 0.681 0.866 Set 3 Obs − 0.173 0.131 0.335 0.660 0.897 Pre − 0.068 0.152 0.347 0.683 0.864 Set 4 Obs − 0.162 0.128 0.331 0.658 0.896 Pre − 0.078 0.147 0.356 0.692 0.867 Set 5 Obs − 0.153 0.128 0.340 0.658 0.899 Pre − 0.095 0.148 0.350 0.685 0.865 Set 6 Obs − 0.153 0.132 0.339 0.661 0.898 Pre − 0.084 0.151 0.344 0.682 0.865 Set 7 Obs − 0.158 0.128 0.346 0.661 0.901 Pre − 0.060 0.148 0.349 0.680 0.857 Set 8 Obs − 0.139 0.123 0.344 0.663 0.901 Pre − 0.073 0.147 0.352 0.683 0.870 Set 9 Obs − 0.151 0.126 0.351 0.665 0.902 Pre − 0.086 0.151 0.345 0.681 0.862 Set 10 Obs − 0.159 0.126 0.340 0.665 0.899 Pre − 0.077 0.151 0.343 0.683 0.865 Mean across 10 sets Obs − 0.156 0.127 0.340 0.662 0.899 Pre − 0.077 0.149 0.347 0.683 0.864 Mean Error (ME) across 10 sets − 0.079 − 0.022 − 0.007 − 0.021 0.035 Fig. 2 The cumulative percent- age of observed EQ-5D-3L util- ity values vs. simulated values for a typical model (model 4/10) 1114 Quality of Life Research (2024) 33:1107–1119 calculate utility as opposed to a simulated distribution with the Monte Carlo simulation method (Supplementary File 1, Table S6). We found that SDs, MSE and MAE values were even less for the Expected utility method than those found for the Monte Carlo simulation method (Table 2) [30]. However, 7.2% of the sample had a utility value less than “0” (worse than death on EQ-5D scale) and 21.4% of the sample had a utility value of “1” (perfect health on the EQ-5D scale), but these values could not be predicted using the Expected utility method. The final algorithm was derived from the entire sample of 4228 family members using Monte Carlo simulation. Details of the final-fitted mlogit models using data from the 4228 family members are given in Table 4. This Table shows the regression coefficients, which can be used to estimate the probability that a respondent will select a particular level of response to questions in the EQ-5D, using individual ques- tion responses from the FROM-16 as predictors. The coef- ficients for individual variables in this type of model are not straightforward to interpret but the results indicate that the FROM-16 question responses most likely to be related to EQ-5D question responses are generally highly significant; for example, responses to the FROM-16 ‘effect on family activities’ question are significantly related to the responses to the EQ-5D ‘usual activities’ question but not to responses to the EQ-5D ‘mobility’, ‘self-care’, ‘pain/discomfort’ or ‘anxiety’ questions; similarly, the FROM-16 responses to the ‘effect on family relationship’ question are significantly related to the responses to the EQ-5D ‘anxiety’ question. This is understandable as the impact on a relative of caring, with its effect on family relationships, could contribute to a relative’s anxiety and depression. An example showing how to use the FROM-16 mapping algorithm is provided in Supplementary File 2. Discussion This mapping of a generic family QoL measure to EQ-5D, facilitates conversion of family member and/or informal carer’s QoL scores into utility values for health economic evaluation. Over six million family members in the UK care for relatives with health conditions [31], with major impact on their QoL [32–35]. However, a major gap in the inclusion of family members in utility analysis may be caused by lack of family member/informal carer data [4]. As value-sets exist for Carerqol-7D and carers’ utility can be assessed directly, perhaps CarerQol-7D use could be prioritised. However, CarerQol-7D, a care-related QoL measure, encompasses dimensions such as support and fulfilment and therefore its scores cannot be summated with patient utilities derived from a health-related utility measure such as EQ-5D-3L [8]. As there is no carer equivalent to EQ-5D, NICE has used EQ-5D to measure carer utility, however, EQ-5D may be inappropriate for family member/informal carers [7]. For example, the EQ-5D question on ‘mobility as a moderate effect’ may mean to family members an inability to go out to meet people, while ‘mobility as an extreme effect’ may confuse family caregivers as to why they should be ‘con- fined to bed’. EQ-5D asks general questions and not specific Fig. 3 Histogram displaying the difference between predicted and observed health utility estimates for a typical cross validation set 4 1115Quality of Life Research (2024) 33:1107–1119 Ta bl e 4 F in al m od el c oe ffi ci en ts (s ta nd ar d er ro rs ) f or e ac h EQ -5 D d om ai n ag ai ns t 1 6 ite m s o f F RO M -1 6, a ge a nd g en de r u si ng m ul tin om ia l l og ist ic re gr es si on FR O M -1 6 ite m s r ep re se nt ed a s F RO M _i te m n am e; N o pr ob le m is th e co m pa ris on g ro up a G en de r w as c od ed m al e = 0, fe m al e = 1 *S ig ni fic an t a t 5 %M ob ili ty Se lf- ca re U su al a ct iv iti es Pa in /d is co m fo rt A nx ie ty /d ep re ss io n So m e pr ob le m s Ex tre m e pr ob - le m s So m e pr ob le m s Ex tre m e pr ob - le m s So m e pr ob le m s Ex tre m e pr ob - le m s So m e pr ob le m s Ex tre m e pr ob - le m s So m e pr ob le m s Ex tre m e pr ob - le m s In te rc ep t −  3 .4 68 (0 .2 2) * −  7 .4 88 (0 .9 62 )* −  3 .2 5 (0 .2 8) * −  6 .4 97 (0 .7 32 )* −  2 .7 46 (0 .2 06 )* −  5 .3 44 (0 .4 22 )* −  2 .1 8 (0 .1 91 )* −  4 .4 37 (0 .3 57 )* −  1 .4 34 (0 .2 05 )* −  4 .0 12 (0 .3 77 )* A ge 0. 03 1 (0 .0 03 )* 0. 00 8 (0 .0 11 ) 0. 00 4 (0 .0 04 ) 0. 00 0 (0 .0 08 ) 0. 01 9 (0 .0 03 )* 0. 00 9 (0 .0 05 ) 0. 02 6 (0 .0 03 )* 0. 01 7 (0 .0 05 )* −  0 .0 13 (0 .0 03 )* −  0 .0 32 (0 .0 05 )* G en de ra −  0 .0 5 (0 .0 79 ) −  0 .0 59 (0 .3 32 ) −  0 .2 97 (0 .1 05 )* 0. 15 3 (0 .2 62 ) −  0 .0 89 (0 .0 79 ) −  0 .2 05 (0 .1 48 ) −  0 .0 08 (0 .0 75 ) −  0 .2 47 (0 .1 31 ) 0. 26 9 (0 .0 83 )* −  0 .0 37 (0 .1 43 ) FR O M _W or - rie d 0. 27 6 (0 .0 78 )* −  0 .1 4 (0 .3 47 ) 0. 38 7 (0 .1 1) * −  0 .2 27 (0 .2 6) 0. 17 (0 .0 76 )* 0. 25 5 (0 .1 58 ) 0. 17 (0 .0 72 )* 0. 38 3 (0 .1 38 )* 0. 52 2 (0 .0 77 )* 0. 71 5 (0 .1 51 )* FR O M _A ng ry 0. 09 4 (0 .0 63 ) 0. 28 8 (0 .2 36 ) 0. 00 9 (0 .0 82 ) 0. 22 1 (0 .1 82 ) 0. 04 2 (0 .0 62 ) 0. 13 7 (0 .1 1) 0. 08 1 (0 .0 61 ) 0. 12 9 (0 .1 01 ) 0. 26 9 (0 .0 69 )* 0. 39 5 (0 .1 07 )* FR O M _S ad −  0 .3 03 (0 .0 67 )* 0. 34 6 (0 .3 34 ) −  0 .1 36 (0 .0 92 ) 0. 62 4 (0 .2 54 )* −  0 .2 11 (0 .0 66 )* 0. 26 4 (0 .1 38 )* −  0 .3 28 (0 .0 63 ) −  0 .3 52 (0 .1 16 )* 0. 11 7 (0 .0 68 ) 0. 35 3 (0 .1 29 )* FR O M _F ru s- tra te d −  0 .0 71 (0 .0 67 ) −  0 .0 78 (0 .3 07 ) −  0 .0 24 (0 .0 92 ) −  0 .5 25 (0 .2 2) * −  0 .0 42 (0 .0 65 )* −  0 .2 66 (0 .1 3) * 0. 02 5 (0 .0 62 ) 0. 01 (0 .1 15 ) −  0 .0 21 (0 .0 67 ) −  0 .0 85 (0 .1 24 ) FR O M _T al ki ng th ou gh ts 0. 19 9 (0 .0 55 )* 0. 45 8 (0 .2 3) * 0. 13 2 (0 .0 73 ) 0. 17 2 (0 .1 68 ) 0. 20 3 (0 .0 53 ) 0. 19 2 (0 .0 99 )* 0. 13 4 (0 .0 52 )* 0. 32 3 (0 .0 9) * 0. 40 3 (0 .0 58 )* 0. 77 7 (0 .0 97 )* FR O M _D iffi - cu lty c ar in g 0. 00 2 (0 .0 69 ) 0. 34 1 (0 .3 04 ) 0. 06 4 (0 .0 91 ) 0. 68 (0 .2 44 )* 0. 09 (0 .0 66 ) 0. 30 1 (0 .1 3) * −  0 .1 21 (0 .0 64 )* 0. 08 5 (0 .1 14 ) 0. 02 9 (0 .0 71 ) 0. 22 3 (0 .1 2) FR O M _T im e fo r s el f −  0 .0 66 (0 .0 7) 0. 20 4 (0 .2 77 ) −  0 .1 59 (0 .0 93 ) 0. 12 4 (0 .2 19 ) −  0 .0 38 (0 .0 68 ) 0. 22 8 (0 .1 27 ) 0. 02 2 (0 .0 67 ) −  0 .2 67 (0 .1 13 )* 0. 24 4 (0 .0 75 )* 0. 35 6 (0 .1 2) * FR O M _T ra ve l 0. 15 2 (0 .0 61 )* 0. 01 8 (0 .2 07 ) 0. 18 7 (0 .0 78 )* 0. 18 3 (0 .1 64 ) 0. 16 1 (0 .0 6) * 0. 31 3 (0 .1 )* 0. 09 3 (0 .0 6) 0. 30 5 (0 .0 96 )* 0. 10 2 (0 .0 72 ) 0. 15 4 (0 .1 05 ) FR O M _E at in g ha bi ts 0. 22 1 (0 .0 61 )* 0. 15 (0 .2 18 ) 0. 29 2 (0 .0 78 )* 0. 14 1 (0 .1 68 ) 0. 10 7 (0 .0 6) 0. 20 2 (0 .1 03 )* 0. 09 4 (0 .0 6) 0. 36 4 (0 .0 95 )* 0. 19 8 (0 .0 71 )* 0. 61 3 (0 .1 04 )* FR O M _F am ily ac tiv iti es 0. 08 3 (0 .0 71 ) 0. 21 1 (0 .3 28 ) 0. 03 1 (0 .0 96 ) 0. 14 9 (0 .2 56 ) 0. 19 (0 .0 68 )* 0. 19 4 (0 .1 41 ) 0. 09 2 (0 .0 66 ) 0. 03 5 (0 .1 2) 0. 1 (0 .0 72 ) 0. 05 9 (0 .1 28 ) FR O M _H ol i- da y −  0 .0 04 (0 .0 63 ) 0. 24 9 (0 .2 87 ) −  0 .0 8 (0 .0 86 ) 0. 27 5 (0 .2 28 ) 0. 02 7 (0 .0 61 ) 0. 08 3 (0 .1 24 ) 0. 00 2 (0 .0 59 ) −  0 .0 27 (0 .1 07 ) 0. 04 7 (0 .0 65 ) −  0 .2 14 (0 .1 12 )* FR O M _S ex li fe 0. 09 5 (0 .0 47 )* −  0 .0 6 (0 .1 77 ) 0. 11 9 (0 .0 63 )* 0. 00 0 (0 .1 4) 0. 10 9 (0 .0 46 )* 0. 20 2 (0 .0 85 )* 0. 11 6 (0 .0 45 )* 0. 13 1 (0 .0 76 ) 0. 06 9 (0 .0 51 ) 0. 07 3 (0 .0 82 ) FR O M _W or k or st ud y −  0 .2 41 (0 .0 65 )* −  0 .1 52 (0 .2 09 ) −  0 .2 62 (0 .0 82 )* 0. 22 7 (0 .1 68 ) 0. 03 7 (0 .0 64 ) 0. 03 2 (0 .1 03 ) −  0 .1 33 (0 .0 64 )* −  0 .3 29 (0 .0 99 )* 0. 03 1 (0 .0 77 ) −  0 .0 28 (0 .1 07 ) FR O M _F am ily re la tio ns hi ps −  0 .0 72 (0 .0 65 ) 0. 45 5 (0 .2 5) 0. 06 8 (0 .0 85 ) 0. 08 1 (0 .1 85 ) −  0 .0 05 (0 .0 64 ) 0. 04 4 (0 .1 11 ) 0. 07 6 (0 .0 62 ) −  0 .0 63 (0 .1 03 ) 0. 21 (0 .0 71 )* 0. 27 7 (0 .1 09 )* FR O M _F am ily ex pe ns es 0. 33 3 (0 .0 58 )* 0. 47 9 (0 .2 31 )* 0. 34 6 (0 .0 77 )* 0. 56 3 (0 .1 82 )* 0. 3 (0 .0 56 )* 0. 39 5 (0 .1 03 )* 0. 23 9 (0 .0 55 )* 0. 52 7 (0 .0 95 )* −  0 .0 83 (0 .0 63 ) 0. 08 2 (0 .1 ) FR O M _S le ep 0. 12 7 (0 .0 65 ) −  0 .1 6 (0 .2 8) 0. 22 3 (0 .0 89 )* −  0 .2 58 (0 .2 13 ) 0. 19 5 (0 .0 63 )* 0. 13 3 (0 .1 26 ) 0. 15 1 (0 .0 61 )* 0. 59 8 (0 .1 13 )* 0. 40 4 (0 .0 67 )* 0. 70 4 (0 .1 2) * 1116 Quality of Life Research (2024) 33:1107–1119 questions about the QoL impact of caring, such as on sleep, relationships and expenses. However, EQ-5D can still be used to assess family member/informal carer utility with some validity [11, 12]. FROM-16, based on the perspec- tive of family members/partners of patients from 26 medical specialities [13] could be used for assessing family member/ informal carer utility when EQ-5D data is unavailable. Per- haps measuring family member/informal carer impact might “double count” QoL impact, but effect on family members is a real additional impact [36]. The study used the method employed [25] for mapping DLQI scores to EQ-5D utility values and followed guidance concerning mapping to obtain EQ-5D utility values for use in NICE health technology assessments [26]. The study used the response mapping method to map FROM-16 responses to EQ-5D using multinomial logistic regression to predict probabilities and the Monte Carlo simulation method to gen- erate predicted EQ-5D responses, the method first used to map SF-12 responses and EQ-5D utility values [15]. In our modelling, we used FROM-16 item scores as continuous independent variables. To have included FROM-16 items with categories may have resulted in only marginal improve- ments, given the complexity of running that model. Further- more, it is not unusual to use item scores rather than catego- ries as independent variables [25]. The mean observed utility across the ten validation sets was 0.67 (SD = 0.33), and the mean predicted utility was 0.66 (SD = 0.27), both consider- ably lower than the UK general population utility value of 0.83 (SD = 0.32) [37]. Since the sample was taken from fam- ily members of patients across > 200 different health condi- tions, this predicted utility already indicates the considerable QoL impact on the patients’ family members/partners. As data were collected during the COVID pandemic, difficul- ties experienced by family members caring for their relative might have contributed to lower utility values. However, our aim was to create equivalence to EQ-5D utility values rather than estimating burden. Most (65%) family members/part- ners were female, representative of the UK gender distribu- tion of carers (68% females) [31, 38]. In this study, the mean difference between observed and predicted utility across ten validation sets was 0.015, indicating a slight but clinically unimportant overesti- mate of poor health. The MSE across ten validation sets ranged from 0.132 to 0.141 (average = 0.137), and the MAE ranged from 0.266 to 0.275 (average = 0.269). Although the mean errors MSE and MAE were slightly higher than in the DLQI mapping study [25] (MSE = 0.073–0.082, mean across 10 sets = 0.077; MAE = 0.187–0.201; mean across 10 sets = 0.193), we are modelling a family-specific measure to EQ-5D, hence such variation is expected. Compared to direct methods, the response mapping method is penalised for any incorrect prediction leading to increased MSE [19, 39]. The model reliably predicts EQ-5D scores, especially at group level, demonstrated through a split-half cross-valida- tion process resulting in very close health utility estimate predictions. On average, 54% of the individual utility dif- ferences were predicted to lie within 0.05 of the actual val- ues: this is comparable to Gray et al.’s findings [19]. 59.12% were predicted to lie within 0.1, 73% within 0.2 and 83% were within 0.3 of actual values. These are still important differences on a scale of 0–1, but the model’s group-level performance demonstrates better predictive ability. Overall predictions were strongly correlated to the observed scores at a group level, the model’s predicting power at individual level requires further evaluation. Other mapping studies with similar results [15, 25] have recommended interpret- ing results at a group level. For successful mapping, there should be conceptual over- lap between the source and target instruments [40]. There were significant correlations between the FROM-16 domains and EQ-5D domains, with emotional domain strongly cor- related to anxiety/depression followed by activity, self- care, pain, and mobility. The personal and social domain of FROM-16 was also strongly correlated to anxiety/depres- sion, followed by activity, pain/discomfort, self-care, and mobility. If an external dataset is not available to assess perfor- mance of a predicted model, random splitting of the sample into an estimation sample and validation sample is recom- mended. This does not result in true randomisation and may result in statistical bias if data is only split once [25]. The Split half-cross validation method [20] used in this study overcomes this disadvantage, improves the accuracy of the model and demonstrates that the predicted utility values accuracy is not due to chance [25]. This method may reduce the sample size of the estimation sample leading to reduced precision. Although our sample was large enough not to be affected by splitting of data, the final model algorithm was based on the entire data sample from 4228 family members/ partners [26]. As our sample came from a UK population of family members/partners across 27 medical specialities and a wide range of condition severities, we believe our model is generalisable to the UK population. We used the response mapping approach which follows the EQ-5D logic by predicting health states and attaching utility tariff values to these. This allows predicted response values to be used in different countries using a country- specific tariff, important as values derived from a UK value set tend to be lower than for other countries [39]. Cultural attitudes might influence HRQoL and utility responses, but a model created on an Italian population worked equally well on a Norway population [25]. When mapping between measures, lack of accuracy in data and lack of test–retest reliability may result in bias. Use in analyses of incremental treatment effects increases the risk of 1117Quality of Life Research (2024) 33:1107–1119 making wrong recommendations about the cost-effectiveness of treatments. This can be minimised by measure developers applying appropriate reliability tests. FROM-16 is responsive to changes in family members’/partners’ health-related QoL over time [41], indicating that it can reliably measure changes in family members’ QoL. Although mapping of FROM-16 to EQ-5D has shown that FROM-16 can reliably predict EQ-5D scores, we do not have evidence that mapping would produce better estimates. Using utility values generated through map- ping is most appropriate when EQ-5D data is not available, as applied by NICE [42]. This study has several strengths. It is the first to explore the relationship between EQ-5D and FROM-16. Although EQ-5D has been mapped to patient generic measures [19], and disease specific measures [15, 21, 39, 43–45], this is the first to map EQ-5D to a family specific measure. The data in this study are representative of family members caring for their relative across all areas of medicine. To justify including carer HRQoL in economic evaluation, the health condition should be associated with a substantial impact on a caregiver’s health and well-being [46]. Caregiver QoL should be assessed using the EQ-5D to be consistent with patient QoL data and to enable comparisons between apprais- als [10, 46]. This study demonstrates that FROM-16 could be an excellent measure to capture this data and associated EQ-5D utilities across all health conditions. There are study limitations. Firstly, no external sample dataset was available for external validation, as unlike patient reported outcomes [47], family outcomes are not regularly measured. Therefore, even though this study demonstrated how well the model performs outside of the sample, external validation with a different dataset of family members is recom- mended. If resources are available, and family members will- ing, FROM-16 and EQ-5D data should be collected directly, though mapping may sometimes be required. The robustness of the mapping model proposed should be further validated in long-term studies. Inclusion of wider socio-demographic variables might improve the models’ predictive performance, but give only marginal improvements, not outweighing the complexity of running the model [19]. Our study sample included a high proportion of family members of patients with neurological conditions: this may have resulted in bias. Accessible versions of our algorithms in a Microsoft Excel spreadsheet with pre-programmed formulae to enable EQ-5D domain probability calculations and health utility estimates from responses to FROM-16 are available on request from the authors. Conclusions Although inclusion of evaluation of burden of family mem- bers/informal carers is encouraged by health economists and decision scientists, this seldom happens, primarily due to lack of family member/informal carer data. Our study fills this important research gap by mapping EQ-5D utility values to FROM-16, a generic instrument which can be used across all disease areas to measure the impact of patients’ disease on their family members’ QoL. The algorithm developed can be used by decision scientists and researchers to calculate EQ-5D-3L utility values from the FROM-16 scores when EQ-5D data is not available, thus allowing the inclusion of the value of the impact on the QoL of family members/infor- mal carers in health technology appraisal. Supplementary Information The online version contains supplemen- tary material available at https:// doi. org/ 10. 1007/ s11136- 023- 03590-z. Acknowledgements We are very grateful to the patient’s family mem- bers and partners who agreed to take part in this study. We wish to thank the following 58 UK-based patient support groups for their help with participant recruitment: (Action for Pulmonary Fibrosis, Against Breast Cancer, Alopecia UK, Arthritis Action UK, Pancreatic Cancer UK, Breast Cancer Now, Carer Wales, Cerebral Palsy CP (SCOPE), CHSS (Chest, Heart and Stroke Scotland), Crohn’s & Colitis UK, Dia- betes UK, Diabetes UK Sheffield, Epilepsy Action, Epilepsy Society, Fibromyalgia and chronic pain support group, Fight for Sight, Genetic Alliance UK, Glaucoma UK, Huntington’s disease Association, IBS patient support group, JDRF.org.uk, Kidney Patient Involvement Net- work, Kidney Wales, Lymphoma Action, MDS patient support group, ME Research UK, Melanoma Action and Support Scotland, Meningitis Now, Migraine Trust, Motor Neurone Disease Association, MS reg- istry, MS Trust, Myeloma UK, National Eczema Society, Ovacome- ovarian cancer charity, Pain Concern, Metabolic support UK, Parents of children with Type 1 Diabetes- Facebook Group, Parkinson’s UK, Pernicious Anaemia Society, Polycystic Kidney Disease Charity, Pro- gressive Supranuclear Palsy Association, Prostrate Cancer UK, Psoria- sis Association, Pulmonary Hypertension Association UK, Retina UK, Royal National Institute for the Deaf, FND Hope UK, Spinal Muscular Atrophy UK, the Asthma UK and British Lung Foundation Partnership, the Brain Tumour Charity, the Encephalitis Society, the National Rheu- matoid Arthritis Society, the Patients Association, the Psoriasis and Psoriatic Arthritis Alliance, UK Parents of Kids with IBD (Ulcerative Colitis and Crohns), Urostomy Association, Verity PCOS UK). We also wish to thank the research support platforms (Healthwise Wales-HWW, Autism Research Centre-Cambridge University database -ARC, Join Dementia Research-JDR) and the social service departments in Wales for their support with participant recruitment for our study. The family/ partner recruitment of this study was also facilitated by HealthWise Wales, the Health and Care Research Wales initiative, which is led by Cardiff University in collaboration with SAIL, Swansea University. Author contributions All authors contributed to the study’s conception and design. Material preparation, data collection and analysis were performed by RS under the supervision of AYF, SMS, SJN and JRI. RK and KO reviewed statistical analysis. RS wrote the first draft of the manuscript, and all authors commented on previous versions of the manuscript. All authors read and approved the final manuscript. Funding There was no external funding for this Cardiff University study. https://doi.org/10.1007/s11136-023-03590-z 1118 Quality of Life Research (2024) 33:1107–1119 Data availability The data are available from the authors on reason- able request from authors according to Cardiff University regulations. Declarations Competing interests RS, KO, RK and SJN declared no competing interest; FMA is employed by Cardiff University: Cardiff University receives royalties from some uses of FROM-16; AYF reports personal fees from Novartis Lecture honorarium, personal fees from Medscape Podcast honorarium, personal fees from Eli Lilly Lecture honorarium, outside the submitted work; MSS and AYF are joint copyright owner of the FROM-16 for which Cardiff University receives royalties for its commercial use, and they receive a share under standard university policy. JRI receives a stipend as Editor-in-Chief of the British Journal of Dermatology and an authorship honorarium from UpToDate. He is a consultant for Abbvie, Boehringer Ingelheim, ChemoCentryx, Citryll, Novartis and UCB Pharma and has served on advisory boards for Ins- med, Kymera Therapeutics and Viela Bio. He is co-copyright holder of HiSQOL, Investigator Global Assessment and Patient Global Assess- ment instruments for HS. His department receives income from royal- ties for the use of the Dermatology Life Quality Index (DLQI), Family Reported Outcome Measure (FROM-16) and related instruments. Ethical approval All participants gave informed electronic consent to participate in the study. Consent to participate Ethics approval was granted by the Cardiff Uni- versity School of Medicine Research Ethics Committee (SREC: 21/19). Consent for publication No details of individual participants are described in this paper. Open Access This article is licensed under a Creative Commons Attri- bution 4.0 International License, which permits use, sharing, adapta- tion, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/. References 1. Pennington, B. M. (2020). Inclusion of carer health-related quality of life in national institute for health and care excellence apprais- als. Value Health, 23(10), 1349–1357. https:// doi. org/ 10. 1016/j. jval. 2020. 05. 017 2. Brouwer, W. B. F. (2019). The inclusion of spillover effects in economic evaluations: Not an optional extra. PharmacoEconom- ics, 37(4), 451–456. https:// doi. org/ 10. 1007/ s40273- 018- 0730-6 3. Basarir, H., Brockbank, J., Knight, C., & Wolowacz, S. (2019). PNS258 the inclusion of the utility values for carers and family members in HTAS: A case study of recent NICE appraisals in the UK. Value in Health, 22, S330. https:// doi. org/ 10. 1016/j. jval. 2019. 04. 1613 4. (ICER), I. f. C. a. E. R. (2019). Spinraza® and Zolgensma® for spinal muscular atrophy: Effectiveness and value: Final evi- dence report. Retrieved September 21, 2023, from https:// icer. org/ wp- conte nt/ uploa ds/ 2020/ 10/ ICER_ SMA_ Final_ Evide nce_ Report_ 110220. pdf 5. Leech, A. A., Lin, P. J., D’Cruz, B., Parsons, S. K., & Lavelle, T. A. (2023). Family spillover effects: Are economic evaluations misrepresenting the value of healthcare interventions to society? Applied Health Economics and Health Policy, 21(1), 5–10. https:// doi. org/ 10. 1007/ s40258- 022- 00755-8 6. Whitehead, S. J., & Ali, S. (2010). Health outcomes in economic evaluation: The QALY and utilities. British Medical Bulletin, 96(1), 5–21. https:// doi. org/ 10. 1093/ bmb/ ldq033 7. Al-Janabi, H., Flynn, T. N., & Coast, J. (2011). QALYs and car- ers. PharmacoEconomics, 29(12), 1015–1023. https:// doi. org/ 10. 2165/ 11593 940- 00000 0000- 00000 8. Hoefman, R. J., van Exel, J., & Brouwer, W. B. F. (2017). Measuring care-related quality of life of caregivers for use in economic evaluations: CarerQol tariffs for Australia, Germany, Sweden, UK, and US. PharmacoEconomics, 35(4), 469–478. https:// doi. org/ 10. 1007/ s40273- 016- 0477-x 9. Al-Janabi, H., Flynn, T. N., & Coast, J. (2011). Estimation of a preference-based carer experience scale. Medical Decision Making, 31(3), 458–468. https:// doi. org/ 10. 1177/ 02729 89x10 381280 10. Wittenberg, E., James, L. P., & Prosser, L. A. (2019). Spillover effects on caregivers’ and family members’ utility: A systematic review of the literature. PharmacoEconomics, 37(4), 475–499. https:// doi. org/ 10. 1007/ s40273- 019- 00768-7 11. McLoughlin, C., Goranitis, I., & Al-Janabi, H. (2020). Validity and responsiveness of preference-based quality-of-life measures in informal carers: A comparison of 5 measures across 4 condi- tions. Value Health, 23(6), 782–790. https:// doi. org/ 10. 1016/j. jval. 2020. 01. 015 12. NICE (2020). CHTE-2020-Health-related quality. Task and finish group report. Retrieved October 8, 2023, from https:// rees- france. com/ wp- conte nt/ uploa ds/ 2020/ 12/ 2020- CHTE- 2020- Health- relat ed- quali ty- of- life-. pdf 13. Golics, C. J., Basra, M. K., Finlay, A. Y., & Salek, S. (2014). The development and validation of the Family Reported Outcome Measure (FROM-16)© to assess the impact of disease on the part- ner or family member. Quality of Life Research, 23(1), 317–326. https:// doi. org/ 10. 1007/ s11136- 013- 0457-y 14. Chantarasap, P., Johns, N. P., Pairojkul, S., Sookprasert, A., Wirasorn, K., Cheawchanwattana, A., Salek, S., & Subongkot, S. (2019). Validation of the Thai version of the family reported outcome measure (FROM-16)© to assess the impact of disease on the partner or family members of patients with cancer. Health and Quality of Life Outcomes, 17(1), 32. https:// doi. org/ 10. 1186/ s12955- 019- 1091-3 15. Wójcik, E., Reszke, R., Otlewska, A., Matusiak, Ł, Ali, F., Finlay, A., & Szepietowski, J. (2020). Family Reported Outcome Meas- ure—16 (FROM-16): Creation, reliability and reproducibility of the Polish language version. Acta Dermato Venereologica, 100(14), adv00219. https:// doi. org/ 10. 2340/ 00015 555- 3583 16. Oh, H. Y., & Shin, Y. S. (2021). Psychometric properties of the Korean family reported outcome measure for family members of patients with acquired brain injury. Journal of Neuroscience Nursing, 53(6), 256–261. https:// doi. org/ 10. 1097/ jnn. 00000 00000 000619 17. Family Reported Outcome Measure (FROM-16). (2014). School of Medicine, Cardiff University. Retrieved August 8, 2022, from https:// www. cardi ff. ac. uk/ medic ine/ resou rces/ quali ty- of- life- quest ionna ires/ family- repor ted- outco me- measu re 18. Shah, R., Finlay, A. Y., Salek, S. M., Nixon, S. J., Otwombe, K., Ali, F. M., & Ingram, J. R. (2023). Meaning of Family Reported Outcome Measure (FROM-16) severity score bands: A cross- sectional online study in the UK. British Medical Journal Open, 13(3). https:// doi. org/ 10. 1136/ bmjop en- 2022- 066168 http://creativecommons.org/licenses/by/4.0/ https://doi.org/10.1016/j.jval.2020.05.017 https://doi.org/10.1016/j.jval.2020.05.017 https://doi.org/10.1007/s40273-018-0730-6 https://doi.org/10.1016/j.jval.2019.04.1613 https://doi.org/10.1016/j.jval.2019.04.1613 https://icer.org/wp-content/uploads/2020/10/ICER_SMA_Final_Evidence_Report_110220.pdf https://icer.org/wp-content/uploads/2020/10/ICER_SMA_Final_Evidence_Report_110220.pdf https://icer.org/wp-content/uploads/2020/10/ICER_SMA_Final_Evidence_Report_110220.pdf https://doi.org/10.1007/s40258-022-00755-8 https://doi.org/10.1007/s40258-022-00755-8 https://doi.org/10.1093/bmb/ldq033 https://doi.org/10.2165/11593940-000000000-00000 https://doi.org/10.2165/11593940-000000000-00000 https://doi.org/10.1007/s40273-016-0477-x https://doi.org/10.1177/0272989x10381280 https://doi.org/10.1177/0272989x10381280 https://doi.org/10.1007/s40273-019-00768-7 https://doi.org/10.1016/j.jval.2020.01.015 https://doi.org/10.1016/j.jval.2020.01.015 https://rees-france.com/wp-content/uploads/2020/12/2020-CHTE-2020-Health-related-quality-of-life-.pdf https://rees-france.com/wp-content/uploads/2020/12/2020-CHTE-2020-Health-related-quality-of-life-.pdf https://rees-france.com/wp-content/uploads/2020/12/2020-CHTE-2020-Health-related-quality-of-life-.pdf https://doi.org/10.1007/s11136-013-0457-y https://doi.org/10.1186/s12955-019-1091-3 https://doi.org/10.1186/s12955-019-1091-3 https://doi.org/10.2340/00015555-3583 https://doi.org/10.1097/jnn.0000000000000619 https://doi.org/10.1097/jnn.0000000000000619 https://www.cardiff.ac.uk/medicine/resources/quality-of-life-questionnaires/family-reported-outcome-measure https://www.cardiff.ac.uk/medicine/resources/quality-of-life-questionnaires/family-reported-outcome-measure https://doi.org/10.1136/bmjopen-2022-066168 1119Quality of Life Research (2024) 33:1107–1119 19. Gray, A. M., Rivero-Arias, O., & Clarke, P. M. (2006). Estimating the association between SF-12 responses and EQ-5D utility values by response mapping. Medical Decision Making, 26(1), 18–29. https:// doi. org/ 10. 1177/ 02729 89x05 284108 20. Kiadaliri, A., Alava, M. H., Roos, E. M., & Englund, M. (2020). Mapping EQ-5D-3L from the knee injury and osteoarthritis out- come score (KOOS). Quality of Life Research, 29(1), 265–274. https:// doi. org/ 10. 1007/ s11136- 019- 02303-9 21. Gray, L. A., Hernandez Alava, M., & Wailoo, A. J. (2021). Mapping the EORTC QLQ-C30 to EQ-5D-3L in patients with breast cancer. BMC Cancer, 21(1), 1237. https:// doi. org/ 10. 1186/ s12885- 021- 08964-5 22. Hurt, L., Ashfield-Watt, P., Townson, J., Heslop, L., Copeland, L., Atkinson, M. D., Horton, J., & Paranjothy, S. (2019). Cohort profile: HealthWise Wales. A research register and population health data platform with linkage to National Health Service data sets in Wales. British Medical Journal Open, 9(12). https:// doi. org/ 10. 1136/ bmjop en- 2019- 031705 23. Dolan, P., Gudex, C., Kind, P., & Williams, A. (1996). The time trade-off method: Results from a general population study. Health Economics, 5(2), 141–154. https:// doi. org/ 10. 1002/ (sici) 1099- 1050(199603) 5:2% 3c141:: Aid- hec18 9>3. 0. Co;2-n 24. McHugh, M. L. (2018). The SAGE encyclopedia of educational research, measurement, and evaluation. SAGE Publications Inc. https:// doi. org/ 10. 4135/ 97815 06326 139 25. Ali, F. M., Kay, R., Finlay, A. Y., Piguet, V., Kupfer, J., Dal- gard, F., & Sam Salek, M. (2017). Mapping of the DLQI scores to EQ-5D utility values using ordinal logistic regression. Qual- ity of Life Research, 26(11), 3025–3034. https:// doi. org/ 10. 1007/ s11136- 017- 1607-4 26. Longworth, L., & Rowen, D. (2013). Mapping to obtain EQ-5D utility values for use in NICE health technology assessments. Value Health, 16(1), 202–210. https:// doi. org/ 10. 1016/j. jval. 2012. 10. 010 27. Mishra, P., Pandey, C. M., Singh, U., Gupta, A., Sahu, C., & Kes- hri, A. (2019). Descriptive statistics and normality tests for statis- tical data. Annals of Cardiac Anaesthesia, 22(1), 67–72. https:// doi. org/ 10. 4103/ aca. ACA_ 157_ 18 28. Coretti, S., Ruggeri, M., & McNamee, P. (2014). The minimum clinically important difference for EQ-5D index: A critical review. Expert Review of Pharmacoeconomics & Outcomes Research, 14(2), 221–233. https:// doi. org/ 10. 1586/ 14737 167. 2014. 894462 29. Brazier, J. E., Yang, Y., Tsuchiya, A., & Rowen, D. L. (2010). A review of studies mapping (or cross walking) non-preference based measures of health to generic preference-based measures. The European Journal of Health Economics, 11(2), 215–225. https:// doi. org/ 10. 1007/ s10198- 009- 0168-z 30. Le, Q. A., & Doctor, J. N. (2011). Probabilistic mapping of descriptive health status responses onto health state utilities using Bayesian networks: An empirical analysis converting SF-12 into EQ-5D utility index in a national US sample. Medical Care, 49(5), 451–460. https:// doi. org/ 10. 1097/ MLR. 0b013 e3182 07e9a8 31. Foley, N., Powell, A., Clark, H., Brione , P., Kennedy, S., Pow- ell, T., Roberts, N., Harker, R., Francis-Devine, B., & Foster, D. (2022). Informal carers research briefings.files.parliament.uk. [Online] Retrieved March 8, 2023, from https:// resea rchbr iefin gs. files. parli ament. uk/ docum ents/ CBP- 7756/ CBP- 7756. pdf 32. Snyder, A. M., Brandenberger, A. U., Taliercio, V. L., Rich, B. E., Webber, L. B., Beshay, A. P., Biber, J. E., Hess, R., Rhoads, J. L. W., & Secrest, A. M. (2022). Quality of life among fam- ily of patients with atopic dermatitis and psoriasis. Interna- tional Journal of Behavioral Medicine. https:// doi. org/ 10. 1007/ s12529- 022- 10104-7 33. Shah, R., Ali, F. M., Finlay, A. Y., & Salek, M. S. (2021). Family reported outcomes, an unmet need in the management of a patient’s disease: Appraisal of the literature. Health and Quality of Life Outcomes, 19(1), 194. https:// doi. org/ 10. 1186/ s12955- 021- 01819-4 34. Noghani, F., Seyedfatemi, N., Karimirad, M. R., Akbarzadeh, A., & Hasanpour-Dehkordi, A. (2016). Health related quality of life in family caregivers of patients suffering from mental disorders. Journal of Clinical and Diagnostic Research, 10(11), vc05–vc09. https:// doi. org/ 10. 7860/ jcdr/ 2016/ 19671. 8792 35. Golics, C. J., Basra, M. K., Salek, M. S., & Finlay, A. Y. (2013). The impact of patients’ chronic disease on family quality of life: An experience from 26 specialties. International Journal of Gen- eral Medicine, 6, 787–798. https:// doi. org/ 10. 2147/ ijgm. S45156 36. Krol, M., Papenburg, J., & van Exel, J. (2015). Does including informal care in economic evaluations matter? A systematic review of inclusion and impact of informal care in cost-effective- ness studies. PharmacoEconomics, 33(2), 123–135. https:// doi. org/ 10. 1007/ s40273- 014- 0218-y 37. Kind, P., Hardman, G., & Macran, S. (1999). UK population norms for EQ-5D. Centre for Health Economics, University of York, Working Papers. 38. Mott, D. (2022). Including carer quality of life in health technol- ogy evaluation: are we there yet?. Retrieved July 31, 2022, from https:// www. ohe. org/ insig ht/ inclu ding- carer- quali ty- life- health- techn ology- evalu ation- are- we- there- yet/ 39. Rivero-Arias, O., Ouellet, M., Gray, A., Wolstenholme, J., Roth- well, P. M., & Luengo-Fernandez, R. (2010). Mapping the modi- fied Rankin scale (mRS) measurement into the generic EuroQol (EQ-5D) health outcome. Medical Decision Making, 30(3), 341– 354. https:// doi. org/ 10. 1177/ 02729 89x09 349961 40. Longworth L, R. D. (2011). NICE DSU technical support docu- ment 10: the use of mapping methods to estimate health state util- ity values report by the decision support unit. . (pp. 18): Sheffield: Decision Support Unit, ScHARR, University of Sheffield. 41. Shah, R., Finlay, A.Y, Salek., S.M., Nixon, S.J., Otwombe, K., Ali, F.M., Ingram, J.R (2023). Responsiveness to Change of the Fam- ily Reported Outcome Measure (FROM-16). Poster presented at: ISOQOL 30th Annual Conference; 18–21 October, 2023; Calgary, Alberta, Canada. 42. Kearns, B., Ara, R., Wailoo, A., Manca, A., Alava, M. H., Abrams, K., & Campbell, M. (2013). Good practice guidelines for the use of statistical regression models in economic evaluations. PharmacoEconomics, 31(8), 643–652. https:// doi. org/ 10. 1007/ s40273- 013- 0069-y 43. Shafie, A. A., Chhabra, I. K., Wong, J. H. Y., & Mohammed, N. S. (2021). Mapping PedsQL™ Generic Core Scales to EQ-5D-3L utility scores in transfusion-dependent thalassemia patients. The European Journal of Health Economics, 22(5), 735–747. https:// doi. org/ 10. 1007/ s10198- 021- 01287-z 44. Hoyle, C. K., Tabberer, M., & Brooks, J. (2016). Mapping the COPD assessment test onto EQ-5D. Value Health, 19(4), 469– 477. https:// doi. org/ 10. 1016/j. jval. 2016. 01. 005 45. Luan, L., Hu, H., & Li, S.-C. (2021). Mapping utility scores from the HeartQoL questionnaire into the EQ-5D for ischemic heart disease. Value in Health Regional Issues, 24, 33–37. https:// doi. org/ 10. 1016/j. vhri. 2020. 02. 001 46. Dawoud, D., Lamb, A., Moore, A., Bregman, C., Rupniewska, E., Paling, T., Wolfram, V., Lovett, R. E. S., & Dent, R. (2022). Capturing what matters: Updating NICE methods guidance on measuring and valuing health. Quality of Life Research, 31(7), 2167–2173. https:// doi. org/ 10. 1007/ s11136- 022- 03101-6 47. Meadows, K. A. (2011). Patient-reported outcome measures: An overview. British Journal of Community Nursing, 16(3), 146–151. https:// doi. org/ 10. 12968/ bjcn. 2011. 16.3. 146 Publisher's Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. https://doi.org/10.1177/0272989x05284108 https://doi.org/10.1007/s11136-019-02303-9 https://doi.org/10.1186/s12885-021-08964-5 https://doi.org/10.1186/s12885-021-08964-5 https://doi.org/10.1136/bmjopen-2019-031705 https://doi.org/10.1136/bmjopen-2019-031705 https://doi.org/10.1002/(sici)1099-1050(199603)5:2%3c141::Aid-hec189>3.0.Co;2-n https://doi.org/10.1002/(sici)1099-1050(199603)5:2%3c141::Aid-hec189>3.0.Co;2-n https://doi.org/10.4135/9781506326139 https://doi.org/10.1007/s11136-017-1607-4 https://doi.org/10.1007/s11136-017-1607-4 https://doi.org/10.1016/j.jval.2012.10.010 https://doi.org/10.1016/j.jval.2012.10.010 https://doi.org/10.4103/aca.ACA_157_18 https://doi.org/10.4103/aca.ACA_157_18 https://doi.org/10.1586/14737167.2014.894462 https://doi.org/10.1007/s10198-009-0168-z https://doi.org/10.1097/MLR.0b013e318207e9a8 https://researchbriefings.files.parliament.uk/documents/CBP-7756/CBP-7756.pdf https://researchbriefings.files.parliament.uk/documents/CBP-7756/CBP-7756.pdf https://doi.org/10.1007/s12529-022-10104-7 https://doi.org/10.1007/s12529-022-10104-7 https://doi.org/10.1186/s12955-021-01819-4 https://doi.org/10.1186/s12955-021-01819-4 https://doi.org/10.7860/jcdr/2016/19671.8792 https://doi.org/10.2147/ijgm.S45156 https://doi.org/10.1007/s40273-014-0218-y https://doi.org/10.1007/s40273-014-0218-y https://www.ohe.org/insight/including-carer-quality-life-health-technology-evaluation-are-we-there-yet/ https://www.ohe.org/insight/including-carer-quality-life-health-technology-evaluation-are-we-there-yet/ https://doi.org/10.1177/0272989x09349961 https://doi.org/10.1007/s40273-013-0069-y https://doi.org/10.1007/s40273-013-0069-y https://doi.org/10.1007/s10198-021-01287-z https://doi.org/10.1007/s10198-021-01287-z https://doi.org/10.1016/j.jval.2016.01.005 https://doi.org/10.1016/j.vhri.2020.02.001 https://doi.org/10.1016/j.vhri.2020.02.001 https://doi.org/10.1007/s11136-022-03101-6 https://doi.org/10.12968/bjcn.2011.16.3.146 Mapping of Family Reported Outcome Measure (FROM-16) scores to EQ-5D: algorithm to calculate utility values Abstract Objective Methods Results Conclusions Introduction Methods Study design and participant recruitment Measures FROM-16 EQ-5D-3L Statistical analysis Exploratory analysis Mapping the FROM-16 to EQ-5D responses Split-half cross-validation Results Study sample demographic characteristics FROM-16 and EQ-5D scores Characterising the distribution and conceptual overlap Split-half cross-validation and model performance Discussion Conclusions Acknowledgements References