Influencing Factors Of Medication Compliance In SLE Patients And Construction And Validation Of Prediction Model

BackgroundSystemic lupus erythematosus（SLE）is an autoimmune-mediated connective tissue disease.Currently,SLE is uncurable,and requires long-term combination medication.However,a multitude of studies have demonstrated poor adherence to medication in SLE patients.Although there are many studies on the factors influencing medication adherence in SLE patients,they are mostly limited to exploring the association between one or several factors and medication adherence,but there is no systematic study on multidimensional adherence as proposed by WHO,and there is also a lack of studies on the construction of predictive models for medication adherence in SLE patients.Therefore,it is necessary to further study the factors influencing medication adherence in SLE patients and construct a multifactorial risk prediction model to solve the problem that individual risk factors are not accurate enough to predict poor medication adherence in SLE patients.Objective（1）To understand the current status of medication adherence in patients with SLE;（2）To reveal the factors influencing medication adherence in patients with SLE;（3）To construct and validate a Prediction model of medication compliance in SLE patients.MethodsA cross-sectional questionnaire study was conducted using a convenience sampling method to select 424 SLE inpatients from the rheumatology departments of several tertiary care hospitals in Anhui Province from February 2021 to December 2021 as the study population.The questionnaire was administered using the general information questionnaire,the disease-related information questionnaire,Self-rating Despression Scale（SDS）,the Systemic Lupus Erythematosus Disease Activity Index（SLEDAI-2000）,Social Support Rating Scale（SSRS）,the Chinese version of the eight-item Morisky Medication Adherence Scale（MMAS-8）,Self-efficacy Appropriate Medication Use Scale（SEAMS）,Self-rated Anxiety Scale（SAS）,The the Visual Analogue Score（VAS）,and Distress Disclosure Index（DDI）were used to investigate SLE patients.Epi Data 3.1 software was used for database creation and data double entry;SPSS 23.0 was used for general data analysis,questionnaire analysis of factors influencing medication adherence and univariate analysis,LASSO regression for variable screening,Logistic regression decision tree and random forest for model construction and validation,from which the best model was selected.Results（1）Basic characteristics of study subjects:The mean score on the Medication Adherence Scale was（4.30±2.51）for 424 patients.More than half of the respondents290（68.4%）had low adherence,69（16.3%）had moderate adherence and 65（15.3%）had high adherence.According to the study,Patients with moderate adherence and high adherence were classified as adherent and those with low adherence as non-adherent using the MMAS-8 cut-off value of 6;the current study population included 68.4%of non-adherent patients and 31.6%of adherent patients.（2）The results of the univariate analysis:Among the socio-economic related factors,occupation,social support and monthly income per capita were not statistically different between the non-adherent and adherent groups（P>0.05）;home location was significantly different between the non-adherent and adherent groups（P<0.05）.Among the health system-related factors,trust in nurses,health care payment method and health education significantly differed between the non-adherent and adherent groups（all P<0.05）.Among the disease-related factors,SAS,SLEDAI-2000,DDI,anti-SSA,U1-RNP,r RNP,VAS,and HGB significantly differed between the non-adherent and adherent groups（all P<0.05）;duration of medication,presence of complications,type of complications,SDS,lupus nephritis,anti-SSB,C₃volume,C₄volume,ESR,ds DNA,CRP significantly differed between the non-adherent and adherent groups（all P<0.05）.There was no difference between the adherence groups and compliance groups（P>0.05）.Among treatment-related factors,SEAMS,reason for current visit,number of admissions,drug efficacy,drug side effects,number of medications,type of medication,method of medication,frequency of medication,medication knowledge,medication management,medication attitude,and family supervision significantly differed between the non-adherent and adherent groups（all P<0.05）;glucocorticoids,HCQ,number of relapses,and medication tools did notdiffer significantly between the non-adherent and adherent groups（all P>0.05）.Among the patient-related factors,gender,education level,forgetfulness,knowledge of the disease,importance of the disease,and importance of the disease by family members significantly differed between the non-adherent and adherent groups（all P<0.05）;age,marital status,BMI,work intensity,housing status,whether or not to smoke,whether or not to drink alcohol,frequency of exercise,and intensity of exercise did not differ significantly between the non-adherent and adherent groups（all P>0.05）.（3）Construction and validation of prediction model:In the Logistic regression,gender,home location,income,frequency of exercise,current visit,HGB,CRP,SLEDAI-2000,VAS,worry about side effects,disease importance and forgetfulness were the predictors of the model;in the decision tree model,worry side effects,current visit,VAS,SLEDAI-2000,SEAMS,occupation,home location and number of hospital admissions were the predictors of In the random forest model,SLEDAI-2000,worry about side effects,SEAMS,literacy,current visit,occupation,disease importance,income,DDI and VAS were the more influential predictors.The AUC value for the Logistic regression model was 0.804,the AUC value for the decision tree model was 0.775,and the AUC value for the random forest model was0.860.The sensitivity and the specificity of the Logistic regression model were0.61,0.77;the sensitivity and the specificity of the decision tree model were 0.61,0.75;the sensitivity of the random forest model was 0.58 and specificity 0.91.The positive predictive value of the Logistic regression model was 0.51 and the correct classification rate 0.73;the decision tree model positive predictive value was 0.49;the correct classification rate 0.71;and the positive predictive value of the random forest model was 0.72;the correct classification rate was 0.82.Conclusion（1）The overall level of medication adherence among SLE patients is relatively high,with 68.4%of them having poor adherence.（2）The factors influencing medication adherence in patients with SLE include socioeconomic-related factors（home location）,health system-related factors（trust in nurses,health care payment methods,and health education）,disease-related factors（SAS,SLEDAI-2000,DDI,anti-SSA,U1-RNP,r RNP,VAS,HGB）,treatment-related factors factors（SEAMS,reason for current visit,number of admissions,drug efficacy,drug side effects,number of medications,type of medication,method of medication,frequency of medication,knowledge of medication,medication management,attitude towards medication,family supervision）,patient-related factors（gender,education level,whether amnesia,knowledge of the disease,importance of the disease,importance of the disease to family members）.（3）The best model in terms of AUC values is the random forest model,which provide a good judgement of medication adherence,with the Logistic regression model being the next best.In terms of specificity,positive predictive value,correct classification rate and AUC value,random forest is the best model.In summary,the random forest model can provide good modelling and prediction of medication adherence.