| Objectives：（1）The objective of the first chapter of the thesis was to analyze and compare theeffect of three complex sampling characteristics (unequal probability sampling, stratumand cluster) on the point estimation and sampling efficiency; and to provide evidence forusing complex sampling adjustment to evaluate the population health measurement data.（2）The objective of the second chapter was to assess the feasibility of the whole aswell as the5individual dimensions of the Chinese version of EQ-5D instrument at itemlevel among common chronic populations using Rasch model.（3） The objectives of the third chapter of this thesis was (a) to grasp the healthstatus characteristics of the residents living in Shaanxi Province as well as the distributionof the health status among different populations; and (b) to examine the main risk factorsof the population health measured by EQ-5D dimensions, EQ-5D index and EQ-5D VASin populations reporting chronic conditions and reporting no chronic conditionrespectively by using complex survey multivariate models.（4）The objective of the fourth chapter was to develop utility-adjusted health service utilization indicators based on health utility, the marginal effects of health utility, healthservice utilization and morbidity; and to apply these indicators to perform cost-utilityanalysis for health service utilization among12populations with common chronic diseases.Methods:（1）The differential representation of the samples was adjusted by combining basicsampling weight, ratio weight and standardization weight. The population was dividedinto6strata according to regional (Shannan, Guanzhong and Shanbei) and urban-ruralclassification. County/district was set as cluster identifier. Three methods were establishedto adjust the complex sampling characteristic. The first method only considered unequalprobability sampling. The second one considered both of unequal probability samplingand the stratum. In the third one, the cluster effect was also included. Finally, wecompared the design effects of the three adjusting methods with simple random samplingmethod and analyzed the effects of unequal sampling, strata and clusters on the parameterestimation and standard error. Four types of estimation were selected, which includedmean, ratio and coefficients in linear regression model and logistic model.（2）The Rasch-based software RUMM2030was applied to analysis EQ-5Dinstrument. The feasibility of the whole and the five dimensions of EQ-5D were evaluatedat item level among a few populations, such as overall population, population whoreporting no chronic conditions as well as12populations reporting common chronicconditions. The methods used to evaluate the whole of EQ-5D included the comparison ofRSM model and PCM model, the overall goodness of fit, the person separation index,targeting and residual principle components analysis. The methods used to evaluate thedimensions of EQ-5D included item residual, threshold and DIF analysis.（3）Univariate statistical ananlysis methods were applied to systematically describethe situation of the health status of overall population in Shaanxi Province, and to examinethe distribution of health status among different populations. Multivariate models weredeveloped to identify the important risk factors of EQ-5D dimensions, EQ-5D index aswell as EQ-5D VAS. The respondents reported chronic conditions and those reported no chronic conditions were analysed independently. Logistic models were adopted toexamine the risk factors of EQ-5D dimensions and Tobit models were adopted to examinethe risk factors of EQ-5D index and VAS. Statistical analysis tools included thecorresponding complex procedures in SAS and STATA. The CSscwmethod constructed inthe first chapter was applied in the statistical analysis.（4）Utility-adjusted health service utilization indicators were constructed based onhealth utility, marginal effect of health utility, health service utilization, morbidity, and thenumber of populations; and were used to perform cost-utility analysis of health serviceutilization. Marginal effect of health utility was estimated using Tobit model. QALY lossresulted from morbidity of chronic diseases were calculated by multiplying the prevalence,the number of population and the marginal effect of health utility. Finally, utility-adjustedhealth service of different chronic populations were calculated and compared with eachothers. Health service utilization indicators included outpatient rate, outpatient expenditure,inpatient rate, inpatient expenditure as well as the length of stay. Populations aged over15,15-59and over60were analyzed independently.Results:（1）After adjusting the differential representation of samples with three kinds ofweight, the proportions of three age groups (20-24,25-29and30-34) increased from4.91%to10.32%,5.23%to7.47%, and4.72%to6.86%respectively. The adjustedage-gender sample structures for the whole province and six strata were consistent withpopulation structures. After the adjustment, the estimations of average health expenditureand prevalence of chronic diseases decreased from932yuan to897yuan, and19.2%to14.2%respectively. The design effects of three complex methods showed that strata,unequal probability and cluster had ascending effects on sampling efficiency. For instance,the design effects of CSw, CSswand CSscwwere1.6,1.58and46.6respectively in theestimation of average of household health expenditure; the design effects of CSw, CSswandCSscwwere1.04,1.04and14.52respectively in the estimation of prevalence rate ofchronic diseases. After the adjustment, some standard errors and P values of the coefficients in multivariate models had increased and some P values became larger thanthe significance level (0.05). For instance, for the CSscwmethod in the linear regressionmodel, the P value of the “region (Shaan Nan)” had increased from <0.0001to0.45, andthe P value of the “employment (employed)” increased from <0.01to0.11; for the CSscwmethod in the Logistic model, the P value of the “region (Guan Zhong)” had increasedfrom0.03to0.31, and the P value of the “alcohol use” increased from <0.001to0.1.（2）All missing rates of five dimensions were smaller than1%.0.13%ofrespondents did not report EQ-5D completely.81.8%of respondents reported perfecthealth status (“11111”) and0.14%of respondents reported worst health status(“33333”). The P values of likelihood ratio tests for PCM model and RSM model inpopulations with chronic conditions were smaller than0.001except for anemia and mentaldisease. The Person Separation Indices of populations with chronic diseases were between0.5and0.8and they were larger than those of population reported no chronic conditionsas well as overall population. The means of individual locations were larger than0, whilethe coefficients of skewness of individual locations were smaller than0. Results ofresidual principle components analysis showed that the eigenvalues of first principlecomponent for all populations were larger than1.4. The proportions of the variance ofprinciple components with eigenvalue of smaller than1.4were between50%and70%.For overall population, population reported no chronic condition, population reportedother chronic conditions, the residuals of five dimensions significantly deviated from therange of-2.5and2.5. The P values were larger than0.05in urban/rural-DIF andgender-DIF analysis for all the12common chronic diseases, while the P value of DIFanalysis was smaller than0.01between populations with and without chronic conditions.（3）(A) For the characteristics of respondents between urban and rural areas,the proportion of respondents who were employed and illiteracies in rural area were22.3and18.1percent larger than those of respondents in urban area respectively; theproportion of respondents with high education level and respondents who never takephysical exercise were10and25.1percent larger than those of respondents in urban arearespectively.(B) For the health situation of overall population, the ranking of percentage of respondents who report health problems for five dimensions were as follows:Pain/Discomfort (10.9%), Depression/Anxiety (5.6%), Mobility (4.6%), Usual Activity(3.9%) and Self-care (2.4%). The medians of EQ-5D index and EQ-5D VAS were0.89and84.51. The Pearson correlation between VAS and EQ-5D was0.47. Pain/Discomfortdimension was significant correlated with VAS and EQ-5D index, where the Pearsoncorrelations were-0.42and-0.89.(C) For the distribution of health status amongdifferent populations, respondents who were female, older, living in north area, living inrural area, reporting health problems during the last two weeks, reporting one or morechronic conditions, widowed or divorced, unemployed, less educated, and lowerhousehold income were more likely to report health problem and lower VAS. All the sevenhealth indicators were distributed as gradient shape among age class, education level,household income class, and the number of chronic conditions. For instance, the EQ-5Dindices of the four age groups were0.899,0.894,0.881and0.848respectively. Thepercentage of reporting health problems in Pain/discomfort dimension for the four agegroups were1.27%、3.56%、6.96%and13.14%.(D) The main risk factors of healthstatus for respondents who reported no chronic conditions included age, educationlevel, and household income. The effects of age class, education level and householdincome class were distributed as gradient shape. For example, the values of Odds Ratio forthree higher education levels in Mobility dimension were3.41,10.23and23.41. Thecoefficients of EQ-5D index were-0.183,-0.355and-0.507respectively.(E) The mainrisk factors of health status for respondents who reported chronic conditions includededucation level, household income and the number of chronic diseases. The effects ofeducation level and household income class were distributed as gradient shape. Forexample, the values of Odds Ratio for three higher education levels in Self-care dimensionwere0.81,0.78and0.56. The coefficients of EQ-5D VAS were2.59,3.63and4.05respectively.（4）(A) In the population aged over15, the prevalence (9.92%) of hypertension wasthe highest among12common chronic diseases; the marginal effect of cancer was thelargest one (-0.3836); musculoskeletal disease resulted the largest QALY loss, which was 828.3years per100thousands persons. However, there were differences for the populationaged60and above. For example, the mental disease resulted the largest marginal effect(-0.3831) and hypertension resulted the highest QALY loss (473years/100thousandspersons).(B) In the population aged15and above, the chronic populations amounted56.6%of outpatients and consumed61.99%of outpatient expenditure; the relative worsehealth status of outpatients included cerebrovascular disease (0.74) and cancer (0.78); bothrate of outpatient (33.1%) within two weeks and the average rate of outpatient expenditure(1045yuan) within two weeks in cancer were the highest comparing with12chronicpopulations; both of the unit utility outpatients(11.1/year) and the unit utility outpatientexpenditure (1342yuan) of cancer were also the highest.(C) The population of inpatients,inpatient expenditure and the length of stay amounted for45%,49%, and51%respectively in the chronic population with the age of15and above. Among thehospitalized populations, the health status of cerebrovascular disease was the worst (0.76);both of the hospitalization rate (33.1%) and the average hospital cost (1045yuan) in cancerwere the highest; both of the unit utility inpatient (0.8/year) and unit utility hospital cost(11882yuan) in caner were the highest; unit utility length of stay (39.4days) in mentaldisease was the highest. There were some differences in the population aged60and over.For example, in the population aged60and over, hospitalization rate (51.7%) incerebrovascular disease was the highest; the length of stay (23.1days) in infectiousdisease was the highest; both of the unit utility inpatient (0.72/year) and the unit utilitylength of stay (28.7days) in cerebrovascular disease were the highest.(D) In thepopulaton aged15and above, the total utility adjusted health service utilization inmusculoskeletal disease was the highest among12common chronic diseases. The cost oftotal utility adjusted outpatients, inpatient expenditure, inpatients, and hospitalization were1.79million,42million,56thousands and1.2billion respectively. There were differencesin the proportions of the total utility adjusted health service utilization between thepopulation aged15-59and population aged over60. The proportions of musculoskeletaldisease, digestive diseases and cancer were much larger in the population aged15-59,whereas the proportions of cerebrovascular disease, hypertension, heart disease anddiabetes were much larger in population aged over60. Conclusions：（1）Unequal probability sampling affected the point estimation while clusteraffected standard error. It is necessary to consider complex sampling characteristics in theevaluation methods for population health. Ignorance of complex sampling characteristicsis likely to result in biased point estimation and incorrect statistical inference.（2）The results of Rasch analysis indicated that the items generally conform to thebasic measurement hypothesis and the fitness of the model was not very well. The mostserious problem was that the instrument was not sensitive enough to relative better healthstatus. Therefore large ceiling effect is likely to occur. In summary, the feasibility ofEQ-5D in populations with chronic conditions is better than that without chronicconditions; and the feasibility of physiological dimensions is netter than that ofpsychological dimensions.（3）The possible risk factors of population health status included age, education,household income and comorbidity. Two out of the four most important risk factors weresocial-economic variables, which reflected the health inequality that resulted fromunbalanced social-economic development. This indicated that we had to promote socialand economical development as well as to improve equality of development.（4）The development of utility-adjusted health service utilization proposed a newmethod for the cost-utility analysis of health service utilization by using data collectedfrom cross-sectional survey. In the population aged15-59, the top6chronic diseases withmore utility-adjusted health service utilization included musculoskeletal disease,hypertension, cancer, cerebrovascular disease, heart disease and digestive disease. In thepopulation aged over60, the top6chronic diseases with more utility-adjusted healthservice utilization included hypertension, heart disease, musculoskeletal disease,cerebrovascular disease, diabetes and digestive disease. |
PDF Full Text Request