Research on the Transformation of Mapping Method for Cancer Patients' Health Utility in the Asia-Pacific Region

Zhang Jie,Sun Quan,Zhang Fang

(School of Business Administration,Shenyang Pharmaceutical University,Shenyang 110016,China)

Abstract Objective To systematically collect the mapping functions of health utility values of various cancer patients in the Asia-Pacific region to provide scientific reference for domestic research in the field of cancer patients' health utility values.Methods A systematic literature search was conducted by using PubMed,ScienceDirect,Web of Science,CNKI,VIP Database and Wanfang Database to collect studies on the application of mapping method for health utility value measurement from 2008 to 2019.The key words included cancer,scale,mapping,and health utility.The inclusion criteria for the studies were:(1)The research papers came from countries in Asia Pacific region;(2)Original research articles;(3)Written/published in Chinese and English.Results and Conclusion A total of 137 Chinese and English articles were retrieved,and 9 articles met the requirements in the screening.The literature was classified by the following types:(1)It had a clear functional relationship;(2)It had no clear functional relationship.Due to the small numbers of samples,the reliability of the research results is not high.The construction of mapping model should build multiple regression models to test the performance of the model combined with different index variables.In addition,due to the limitations of the research objects selected in the literature,more attention should be paid to the mapping function of other types of cancers.At the same time,the research and development of the original data should be focused on as well.

Keywords:cancer patients;scale;mapping;health utility

In the measurement of health utility value,mapping method refers to the transformation of non-preference-based information in the initial scale into utility index through modeling,and the functional formulas of each model are obtained.The interpretation and prediction ability of each model are evaluated by reserved data,and finally the best utility value conversion model is obtained.The basic logic of the mapping method is shown in Fig.1.

At present,the research on health utility mapping model abroad has a certain historicity and scale,which is mainly reflected in the depth and breadth of the research.However,domestic research on health utility mapping model has just started.Fu Xijing,Liang Minhong and Sun Mao et al.introduced the application principle,process and selection of models needed for mapping method in measuring health utility value[1].Liu Tong,Li Shunping and Chen Gang systematically sorted out the core idea,applicable model,application situation and shortcomings of mapping method[2].Sun Yuanyuan,Yu Zheng and Li Hongchao systematically discussed the mapping method and its related models for health utility value.They introduced the application of each model in probability mapping with the example of converting the measurement results of quality of life scale into the utility values of European five-dimensional health scale[3].Zhou Ting and Ma Aixia elaborated on the common empirical models and international application status of disease specific scale mapping[4].John E.Brazier et al.conducted a review of health measurement studies based on non-preference and general preference[5].Many researchers have carried out in-depth analysis from the aspects of research content,progress,tools and status,which reveals the future research direction.

Fig.1 Basic logic of mapping method

The current situation and trend of cancer epidemiology in China are not good.According to the global cancer report from World Health Organization,9.6 million people died of cancer in 2018,equivalent to 1/8 and 1/11 of the deaths of men and women.Compared with the 8.2 million cancer deaths in 2012,this figure is significantly higher,and the number of new cases is also increasing.However,the latest cancer data released by the National Cancer Center in 2018 shows that cancer is the main cause of death for Chinese.With the growth of age,the incidence and mortality rate of cancer for men and women in China are increasing[6].Based on the trend of cancer epidemics and the comprehensive burden of cancer in China,it is of great significance to systematically collect the research on the mapping method for cancer health utility value in Asia Pacific region so that a reasonable and scientific quality of life assessment scale can be used to evaluate and improve the quality of life of specific population[7].

1 Research progress

1.1 Search situation

Taking the ten years from 2008 to 2019 as the searching period and using cancer,mapping and scale as keywords,35 Chinese documents were retrieved from CNKI and Wanfang Database.Taking cancer,mapping and health utility as keywords,we searched PubMed,ScienceDirect and Web of Science,and retrieved 102 English documents.In the analysis of the included articles,each study was classified according to the following types:(1)The studies with explicit mapping function;(2)The studies without explicit mapping function.Both types of studies included the mapping between disease-specific scale and universal scale and the mapping between universal scale and universal scale.137 articles were identified through literature search in Chinese and English databases.After two rounds of screening,9 English literatures that met the screening criteria were finally identified,all of which were empirical articles rather than reviews.The classification of these literatures according to study design is shown in Fig 2.

1.2 Status of scale research

1.2.1 Studies of explicit mapping function

Three papers have given explicit mapping functions.They included not only mapping research between disease-specific scale and universal scale,but also the mapping research between universal scale and universal scale.They were summarized according to the author,year,disease,sample size,regression model used,initial scale,target scale,mapping function,etc.The summary of documents in the schedule is shown in Table 1.

Fig.2 Articles selection process

Table 1 The information for retrieving literature

Askew RL[8]et al.took 273 American patients with melanoma as the study subjects,and took the functional assessment of cancer therapymelanoma(FACT-M)and EuroQol five-dimensional questionnaire(EQ-5D)scores of cancer treated melanoma as explanatory variables,as well as race/ethnicity,age,gender,marital status,and AJCC melanoma stage.The censored least absolute deviation(CLAD)and the ordinary least square(OLS)regression analysis were used to map the model,and the performance of the model was checked byR2,which compared the residuals and verified the fit in the data.The results showed that the OLS mapping function had better prediction ability and could map from FACT-M to EQ-5D practical score.When there was no direct population preference measure,it would be helpful to deduce the utility program.

Fu Xijing[9]et al.collected the information of 676 Chinese lung cancer patients based on the Functional Assessment of Cancer Therapy-Lung(FACT-L)Chinese version(V4.0)and Chinese version of EQ-5D data,then they took the FACT-L scores and age and gender indices as explanatory variables.OLS,the generalized linear model(GLM),Tobit model,CLAD and quantile regression models were used to map the utility value integral systems in China,Japan and Britain.R2,mean absolute error(MAE)and root mean squared error(RMSE)were used as model performance evaluation indicators for comparative analysis.The results showed the mapping model between FACT-L and EQ-5D based on the Chinese population had good predictive ability,and accurately converted the non-preference life quality information of lung cancer patients into health utility values.This is the only empirical mapping study retrieved in China.

Khan I[10]et al.took the questionnaire data of 100 non-small cell lung cancer patients as an example,using age,gender,smoking status,stage,and histology as explanatory variables.The mapping algorithm between the EQ-5D-3L,EQ-5D-5L and European organization for research and treatment of cancer quality of life questionnaire core-30(EORTC QLQ-C30)was determined by using the random effect linear regression model,Beta-binomial(BB)and limited dependent variable mixture model(LDVMM)respectively.The results showed that the BB mapping algorithm could be better applied to EQ-5D-3L and EQ-5D-5L.In addition,EQ-5D-5L could provide better predictions under poor health conditions,while several algorithms previously using EQ-5D-3L were generally over-predicted.

1.2.2 Mapping without explicit transformation formulas

A total of 6 documents did not give a clear mapping function.Cheung[11]et al.used clinical data of 558 Singaporean cancer patients,taking the physical,emotional and functional status dimension scores as explanatory variables.They attempted to establish a mapping model between the English and Chinese versions of the functional assessment of cancer therapy-general(FACT-G)scores and the EuroQoL Group's EQ-5D utility index through OLS and CLAD methods respectively.R2and MAE were used as model performance evaluation indicators for comparative study.The results showed the social and family factors of FACT-G were poorly correlated with the EQ-5D utility index,while the algorithm built by CLAD had better performance and could accurately map the FACT-G(Chinese and English versions)utility values to EQ-5D.

Doble and Lorgelly[12]collected the data of 3 560 Australian cancer patients,using QLQ-C30 total score and its interaction term as explanatory variables to construct 10 mapping algorithms between QLQ-C30 and EQ-5D-3L by OLS and quantile regression(QR).Meanwhile,RMSE and MAE were used as indicators.The results showed two out of the 10 algorithms could construct accurate relationship between QLQ-C30 and EQ-5D-3L,with the indicators of the two models performing well.

Wong[13]et al.collected data of 509 Hong Kong patients with colorectal cancer,and used the cancer quality of life questionnaires(QLQ-C30 and QLQ-CR29)scores and their weighted variables as explanatory variables.Then they constructed mapping models between QLQ-C30 and SF-6D,QLQ-CR29 and SF-6D.The fitting degree of the model was examined by using exploratory power(R2and adjustedR2).The results showed both scale and item response models could explain more than 67% of the variation in SF-6D scores,thus indicating SF-6D scores could be predicted from QLQ-C30 and QLQ-CR38/CR29 scores with satisfactory precision.

From 893 Korean cancer patients,Kim[14]et al.used the clinical data,body,role,mood and pain index of EORTC QLQ-C30 as explanatory variables to construct an OLS multiple linear regression model.RMSE was chosen as the indicator of the performance of the model.The results showed this algorithm could accurately establish the mapping model between QLQ-C30 and EQ-5D,and it could be used to convert the utility value of cancer patients in Korea.

In another study,Kim et al.used the clinical data of 199 Korean patients with metastatic breast cancers as the study samples[15].They used the sub-items of EORTC QLQ-C30 and the European organization for research and treatment of cancer quality of life questionnaire breast cancer-23(EORTC QLQ-BR23)questionnaires as explanatory variables to construct six models through OLS.R2,MAE and RMSE were used as indicators for evaluating model performance of the mapping between QLQ-C30 and EQ-5D,QLQBR23 and EQ-5D.The results showed the regression model with the sub-item score of QLQ-C30 had the best performance and good predictive validity.

Teckle et al.[16]used the FACT-G,EQ-5D and the Short Form-6D(SF-6D)questionnaire scores from 367 Canadian patients with cancer as the regression data to build three mapping models between FACT-G and EQ-5D,FACT-G and SF-6D through OLS,GLM and CLAD.Then RMSE and MAE were applied to predict scale utilities.The results showed the GLM predicted SF-6D scores matching the observed values more closely than the CLAD and OLS.Physical,functional,and emotional well-being domain scores of the FACT-G best explained the EQ-5D and SF-6D,and both mapping models built by GLM performed well.

2 Characteristics of literature research

2.1 Research content

The number of samples of cancer utility health values selected by mapping study was small,and the number of samples in most literature was less than 1 000.Small sample size will lead to lower prediction accuracy.Therefore,the results of the study have large errors and the conclusions are subjective,which can directly lead to lower representativeness and reliability of other patients in one field.Moreover,the inherent differences in population(For example,socio-cultural,health,clinical practice models and access to health care services)may affect some of the explanatory results and limit the externalities of research findings.It is suggested that scholars should take statistical errors,clinical errors and group differences into consideration and select representative research samples to improve the prediction accuracy.

Due to the effects of extreme health conditions,mapping models are often underestimated or overestimated,resulting in inaccurate prediction.Mixed model is less affected,but it does not perform well near the distribution center.In addition,the same observation indicators used in different clinical problems have different performance,and the research conclusions are not very pertinent.This systematic bias may lead to underestimated health benefits,especially the quality of interventions to improve quality of life.Overestimate and systematic bias are common in the most OLS models used.It is suggested that scholars should use multiple models to match different problem situations and apply mapping algorithm cautiously.Sensitivity analysis is also recommended to evaluate the impact of the choice of these algorithms on costbenefit studies[17]..At the same time,it is necessary to increase the types of observation indicators,and carry out multi-faceted studies.The in-depth evaluation of the mapping model can reduce the calculation amount and the phenomenon of high value underestimation or low value overestimation.

Cancer covers hundreds of categories and involves multiple diseases.However,the results of current literature research in the Asia-Pacific region indicate that the types of cancer selected by the mapping method are relatively limited.In China,there is only one empirical study on the transformation of cancer health utility value mapping methods and most other mapping studies are still at the level of secondary analysis and literature inference.It is recommended that scholars broaden the research field and pay more attention to the mapping function of other types of cancer.Besides,they should attach importance to the research and development of raw data for enriching the research.

2.2 Scale

The performance of the mapping model is related to the degree of overlap between the instruments.In the current review,most studies have chosen EQ-5D as the target scale,followed by SF series scale.However,the results of many studies show that the general scale represented by EQ-5D is an important aspect of the target scale that does not fully cover condition-specific indicators[18].For example,EQ-5D does not contain dimensions of energy or vitality.In another example,SF-36 has been shown to have significant floor effects,and EQ-5D has a high ceiling effect[19].

Some scholars take single diseases such as lung cancer,melanoma,colorectal cancer,and breast cancer as research subjects,while others use the data of all cancer patients as the research basis.Clinical trial data is the main data source for the samples applied by mapping method.The selection of the initial scales involves the universal scale and the disease-specific scale.Among them,the disease-specific scale includes FACT-M,FACT-L,FACT-G,QLQ-C30,QLQBR23,QLQ-CR29,and QLQ-CR38.The disease universal scales are EQ-5D,EQ-5D-3L,EQ-5D-5L,and SF-6D,of which EQ-5D is more commonly used.The EQ-5D-5L provides better predictions in poor health conditions.However,several algorithms that previously used the EQ-5D-3L are generally overestimated.

2.3 Econometric methods

Most of the research models constructed with demographic variables and clinical measurement results are used as explanatory variables,and utility index and scores of each dimension as dependent variables.In the selection of econometric methods,OLS has the highest application frequency.CLAD,GLM,Tobit,Beta-Binomial,LDVMM and QR are widely used.Among them,the mapping models established by OLS,CLAD and GLM all can perform well.The accuracy and fitting effect indicators of the model are determined byR2,adjustedR2,comparative residual,and AIC.RMSE and MAE are used to evaluate the predictive ability of the model.The best mapping model is selected by considering the fitting effect index and the forecasting effect index.

3 Discussions

At present,as to the research on cancer mapping method in the Asia-Pacific region,China is still in the preliminary stage.We have only one empirical study to accurately convert the non-preference quality of life information of lung cancer patients into health utility values.However,the relevant empirical research abroad is relatively mature,which has developed a multi mapping algorithm for the transformation of cancer utility health value.In the current review of related research,the most common estimation method is OLS,followed by GLM and CLAD.Some scholars worry that the standard OLS regression model can underestimate the level of uncertainty in the estimation.The mapping model obtained by other methods also has some problems,such as the ability of model interpretation,the coverage of physical size and so on.It is suggested that researchers use different regression methods to construct multiple mapping models and compare the comprehensive performance of each model according to multiple evaluation indicators.In addition,the research scope of disease types should bet comprehensive.However,the mapping studies of many common diseases have not yet been carried out.This reveals that more mapping studies may lead to common diseases that may require knowledge and clinical application.In the later exploration of the application of health utility value,relevant empirical research should be carried out.With the increasing emphasis on the quality of life and the in-depth study of health utility value,research on the scale mapping method of cancer in various countries will develop mature,and the mapping method suitable for China will also be presented more systematically in the future.