Development And Assessment Of General PRO Measure For Cancer Based On Bayesian Generalized Partial Credit Model And Ordinal Bayesian Instrument Development Method

Objective:Based on the characteristics and functions of PRO in measuring disease burden and evaluating clinical efficacy,a general PRO measure of cancer suitable for regional and cultural characteristics in China was developed.The quality of life,therapeutic effects and side effects of drugs were evaluated by obtaining PRO data of patients with cancer.The minimum clinical inportant difference was calculated according to the guidelines issued by FDA in 2006,which was used to explain clinical problems in the changing score of PRO measure.Explore potential subgroups of cancer patients by developing the general PRO measure of cancer and determine these risk thresholds of PRO among different subgroups It is a foundation for the individualization of clinical treatment and nursing intervention of patients with cancerMethods:Based on the draft guidelines of the PRO Scale developed by FDA,by consulting relevant experts and recording relevant information through interviews with cancer patients,the theoretical framework and item pool of the general PRO measure were formed for cancer patients.And the redundant and irrelevant items were deleted through expert scoring and cognitive tests.We debuged the wording of the questionnaire language,added new items,and formed the initial scale.This questionnaire survey was conducted in 7 hospitals of different levels in Shanxi Province and Cancer Hospital of Henan Province First of all,the first item-selection process was analyzed for pre-survey.The items that did not meet the requirements were eliminated,and similar items were merged.And the revised scale was used for formal investigation.In the formal survey,the original version of Cancer-PROM was screened by classical measurement theory after the data was recovered.And then the responsiveness of Cancer-PROM was evaluated in respiratory,digestive,blood and endocrine disease systems by using the generalized partial credit model(GPCM)under the concept of frequency and Bayesian respectively.The related items with poor responsiveness were deleted.The final version of Cancer-PROM was formed.The general cancer scale based on patients' self-report was evaluated by reliability(Cronbach's alpha coefficient and entire reliability),validity(content validity,construct validity,OBID and discrimination validity)and feasibility analysis(acceptance rates,completion rates and response time of these questionnaire).Minimum clinical significance difference(MCID)in each doamin was calculated to predict the survival of cancer patients According to the heterogeneity of cancer patients,latent profile analysis(LPA)was applied to explore potential subgroups.The rationality of subgroups was further verified by some variables in the basic situation and health status of patients with cancer.The kernel density curves of various dimensions in physiological and psychological doamains were drawn.The corresponding risk threshold of PRO was determined by the intersection points of these curves.Results:This study had formed a conceptual framework of four domains(physiological domain,psychological domain,social domain and therapeutic domain)and 13 dimensions(common symptoms,sleep,appetite,pain,anxiety,despair,social impact,social adaptation,the satisfaction of treatment,patients' compliance,side effects of surgical,chemotherapy and radiotherapy).the initial item pool was generated,consisting of 83 items.After expert scoring and cancer patients' cognitive test,we deleted eleven items,and debugged the items linguistically.A total of 2800 questionnaires(2200 cancer patients and 600 non-cancer patients)were distributed in pre-survey and formal survey.Finally,2490 questionnaires(1958 cancer patients and 532 non-cancer patients)were collected,including 591 patients with cancer of pre-survey and 1367 patients with cancer of formal survey.The baseline data of the two groups were compared by t-test and chi-square test.Pre-survey used classical test theory to screen items,and suggested to delete 17items.For the practical significance of these items,five items in physiological domain(PHD1,PHD4,PHD9)were temporarily retained.These items of PHD7,PHD10 and PHD12 were merged into one item(named abnormal gastrointestinal symptoms).Two items in therapeutic domain(TRD8,TRD13)were temporarily retained.A scale of 61 items was formed.On this basis,a formal survey was conducted and CTT are used again to select items.We deleted eight items.Finally the initial version of Cancer-PROM with 53 items was formed.Then items'responsiveness of Cancer-PROM were analyzed according to the data from cancer patients of the respiratory,digestive,blood and endocrine systems.The data included 436 cases of respiratory system,710 cases of digestive system,267 cases of blood system,and 182 cases of endocrine system.The responsiveness of the items on four systems was evaluated by using the generalized partial scoring model(GPCM)under the frequency and Bayesian theory.And four items with poor responsiveness were deleted.The final version of Cancer-PROM(4 domains,13 subdomains and 49 items)Then,the final version of Cancer-PROM was evaluated.The Cronbach's alpha and entire reliability coefficients of each subdomain were greater than 0.7.In the early development stage of the scale,the content validity was guaranteed by literature search,expert consultation and cognitive test of patients.The standardized factor loadings of confirmatory factor analysis were mostly greater than 0.4.These Bayesian estimated value of factor loadings calculated by OBID method was more than 0.4,which indicated that the scale met the expected research purpose.The average response time of cancer patients was 14.2 minutes,indicating the feasibility of the scaleDistribution-based method was used to determine the minimum clinical difference(MCID)in physiological,psychological,and social domains,which were 5.63,3.42,and 4.16 respectively.It was convenient to evaluate the health status of cancer patients.Latent profile analysis(LPA)was used to analyze the heterogeneity of cancer patients.The physiological domain was divided into two subgroups(the high physiological function group and low physiological function group).The risk thresholds of PRO for common symptoms,sleep,diet and pain were 71.74,71.28,66.29 and 65.16,respectively.The psychological domain was divided into two subgroups(high psychological quality group and low psychological quality group).The risk thresholds of PRO in anxiety and disappointment were 59.56 and 66.60,respectivelyConclusion:The general patient-reported outcome measure of cancer(Cancer-PROM)has a good reliability,validity and feasibility,and can be used as an effective clinical evaluation tool for patients with cancer.The MCID and risk thresholds of PRO were determined to provide evidence for assessing the health status and quality of life of patients with cancer.