| BackgroundPremature ovarian insufficiency(POI)is a clinical syndrome of ovarian dysfunction(follicular depletion,decline or loss of endocrine function and reproductive function)in women before the age of 40.Epidemiological evidence shows that the prevalence of POI increased from 1%in 1986 to 3.7%in 2019,a nearly three-fold increase in 33 years.POI seriously affects women’s physical,psychological,and reproductive health,and endangers women’s long-term health after menopause.With the gradual increase in people’s awareness and attention to POI,how to evaluate the disease state and disease progression of POI,and select appropriate treatment methods to improve ovarian function,reproductive outcomes,and quality of life in POI patients has become an increasingly important issue in the reproductive and gynecological circles at home and abroad.Concerns.At present,there is still a lack of optimal intervention strategies for the treatment of POI.Traditional Chinese medicine,especially acupuncture therapy,has attracted increasing attention in the treatment of POI.Acupuncture therapy have been widely used in a variety of ovarian dysfunction diseases.However,acupuncture treatment of POI is mostly reflected in the improvement of ovarian function and quality of life in the short term(3-6 months),and there is a lack of evidence on factors affecting the efficacy of acupuncture,long-term follow-up,and pregnancy outcomes.Pregnancy outcomes in POI patients were poor,with a spontaneous pregnancy rate of 4.4%and a live birth rate of 5.8%after assisted reproductive technology intervention.Because POI patients are relatively young,most of them still have fertility requirements.Compared with perimenopausal symptoms caused by low estrogen,POI patients are more urgent about how to get pregnancy as soon as possible.Therefore,it is particularly important to determine the clinical efficacy of POI patients and predict pregnancy outcomes.The disease of POI is complex,the hazards are diverse,and the follow-up period is long.By conducting real-world study or establishing a POI registration database,and fully and continuously collecting POI-related diagnosis and treatment information,it is possible to manage its health and solve various clinical problems dynamically,comprehensively,and accurately.With the widespread application of artificial intelligence(AI)technology in the medical field,medical clinical decision-making has shifted from being dominated by personal experience,to being based on evidence-based medical evidence,and then to a reliable decision-making model based on clinical data.AI can deeply mine the correlation between medical data,analyze past medical information with the help of machine learning,and build models of disease occurrence,diagnosis,and prognosis evaluation to assist clinical decision-making.ObjectiveThis study intends to carry out a multi-center,large-sample,prospective real-world study through the International Patient Registry Platform of Acupuncture-Moxibustion(IPRPAM)to collect real-world clinical diagnosis and treatment data of acupuncturemoxibustion treatment of POI.With the help of data mining and machine learning methods,the clinical characteristics of POI patients were analyzed and the ovarian function biomarker thresholds for POI progression were determined to form disease progression status assessment and clinical efficacy judgment criteria.To explore the pregnancy outcome,long-term efficacy,beneficiary population and influencing factors of clinical efficacy of acupuncture for POI;and to construct three machine learning prediction models for predicting pregnancy possibility,pregnancy time and clinical efficacy.MethodThis study is based on a large sample,multi-center acupuncture treatment of POI clinical characteristics,disease progression,efficacy prognosis and pregnancy prediction research based on the Chinese population,to provide reliable,continuous and detailed evidence-based medical evidence for the clinical diagnosis and treatment of POI.1 Analysis of clinical characteristics of patients with premature ovarian insufficiency.A cohort study based on real-world data from a multicenter patient registry.Divided into 2 cohorts according to the exposure factor FSH level:subclinical POI(15 IU/L<FSH≤25 IU/L)and POI(25 IU/L<FSH≤40 IU/L),to compare the clinical characteristics of patients with subclinical POI and POI.Using association rules and complex network analysis to form core symptom clusters;constructing a local weighted regression(lowess)smooth curve to fit the changing trends of age and ovarian function indicators;using restricted cubic spline(RCS)based on Cox regression model to divide the cut-off values of ovarian function-related indicators.And the ROC curve was used to compare the accuracy of the combined determination of different ovarian function-related indicators to determine the best combination of indicators for the progression of POI.2 Analysis of the influencing factors of acupuncture and moxibustion on pregnancy and curative effect of patients with premature ovarian insufficiency.A cohort study based on real-world data from a multicenter patient registry was divided into two cohorts according to whether exposure factors were combined with drugs:acupuncture group and acupuncture combination medicine group.The primary efficacy indicator was clinical pregnancy rate,and secondary efficacy indicators were live birth rate,miscarriage rate,menstrual recovery rate,ovarian function indicator response rate,clinical response rate,and patient-reported outcomes.Taking the efficacy index as the target event,the evaluation time point of the efficacy index as the time variable,and all the influencing factors as independent variables,a univariate Cox proportional hazards model(Cox regression)was constructed.Hazard Ratio(HR)and 95%confidence interval(Confidence Interval,CI)were calculated,and the log-rank test was used to compare the differences of different Kaplan-Meier survival curves,with P<0.05 as significant difference.Analyses were stratified by number of acupuncture(≤24 vs.>24),FSH level(≤40 IU/L vs.>40 IU/L),and menstrual status(amenorrhea vs.nonamenorrhea).Stabilized Inverse Probability of Treatment Weighting(IPTW)was used to adjust for confounding factors.3 Construction of a machine learning-based pregnancy prediction model for acupuncture treatment of premature ovarian insufficiency.Supervised machine learning algorithms were used to construct a pregnancy prediction model,a pregnancy time prediction model,and a clinical efficacy prediction model.The best predictor feature combination was comprehensively screened by three methods:correlation analysis,feature engineering method and the FSelectorSequential function in the R language mlr3fselect package.Apply 8 machine learning classification algorithms(Logistic Regression,Random Forest,Naive Bayes,Polynomial Log Linear Neural Networks,Support Vector Machines,Extreme Gradient Boosting,K-Nearest Neighbors and Linear Discriminant Analysis)and 6 regression algorithms(Linear Regression,Random Forest,Kriging Regression,Support Vector Machines,Extreme Gradient Boosting,K-Nearest Neighbors)to develop different machine learning models.According to 7:3,it is divided into training set(Train)and validation set(Test),using leave-one-out cross-validation training model,using the area under the Receiver Operating Characteristic Curve(ROC)curve(AUC),accuracy(ACC),Fp,Logloss,Mean Square Error(MSE)to evaluate model performance,and ROC and PrecisionRecall(PR)curves to determine the most accurate machine learning model.Results1 Analysis of clinical characteristics of patients with premature ovarian insufficiency(1)Etiology and clinical symptomsThere was a statistically significant difference in age between subclinical POI patients and POI patients(33.8 ± 4.0 years vs.32.6 ± 4.6 years,P=0.033).69%(378/548)of POI patients were still unexplained,iatrogenic factors accounted for 18.7%(103/548),psychiatric factors accounted for 6.2%(34/548),and lifestyle factors accounted for 2.4%(13/548).The symptoms of the top ten POI patients are:hot flashes(62%),fatigue(50%),easy fatigue(48%),depression(47%),sleep disturbance(47%),chills(44%),night sweats(42%),infertility(40%),low libido(27%),and vaginal dryness(26%).(2)Menstrual characteristicsThere were statistically significant differences in the age of abnormal menstruation and the age of diagnosis between patients with subclinical POI and those with POI(30.8± 5.1 vs.29.4 ± 5.4 years,P=0.014;33.6 ± 4.1 vs 32.1 ± 4.9 years,P=0.003).There was a statistically significant difference in the reproductive period between subclinical POI patients and POI patients(18[14,22]years).vs 16[12,20]years,P=0.011).The amenorrhea rate and menstrual cycle were significantly different between subclinical POI patients and POI patients(27%vs 51%,P<0.001;50[30,60]days vs 30[28,43]days,P<0.001).(3)Features of ovarian functionAs expected clinically,with POI disease progression,FSH and LH increased,while E2,AMH,and AFC decreased.Contrary to clinical expectations,FSH/LH ratio decreased with POI progression.Compared with subclinical POI patients,POI patients were statistically significant differences in FSH,LH,FSH/LH ratio,E2,AMH and AFC(21[19,23]IU/L vs.55[35,87]IU/L,P<0.001;6[4,9]IU/L vs.27[12,45]IU/L,P<0.001;3.45[2.40,4.61]vs.2.32[1.71,3.12],P<0.001;29[18,46]pg/ml vs.23[12,39]pg/ml,P<0.001;0.27[0.12,0.58]ng/ml vs.0.06[0.02,0.15]ng/ml,P<0.001;3.5[2,5]vs.1[0,2],P<0.001).(4)Cut-off value of ovarian function-related indicatorsReduced risk of POI progression with FSH/LH ratio<2.5,AMH>0.05 ng/ml,AFC≥ 2,LH>17.5 IU/L and E 2>25 pg/ml;FSH/LH ratio>2.5,AMH<0.05 ng/ml,AFC≤1,LH<17.5 IU/L,and E2<25 pg/ml,increased risk of POI progression.The five indicators of FSH/LH+LH+AMH+AFC+E2 combined to determine the progression of POI had the highest AUC:0.869[95%CI:0.834,0.904].2 Influencing factors of acupuncture and moxibustion on pregnancy and curative effect in patients with premature ovarian insufficiencyA total of 36 doctors were recruited from 28 hospitals in 13 provinces to participate in the study.A total of 503 POI patients who met the diagnostic criteria of ESHRE were included in the analysis,including 394(78.3%)in the acupuncture group and 109(21.7%)in the acupuncture combined medicine group.(1)Baseline characteristics503 patients with POI included in the analysis,with an average age of 32.6 ± 4.6 years,67%(336/503)with college education or above,the average age of abnormal menstrual cycle was 29.4 ± 5.4 years,and the median disease duration was 24[12,49]months,amenorrhea patients accounted for 50%(251/503),50%(249/503)had a history of traditional Chinese medicine treatment,42%(211/503)having a history of HRT treatment,and had surgery history accounted for 31%(155/503),family history accounted for 3%(15/503).FSH level was 56[35,87]IU/L,LH level was 28[12,45]IU/L,FSH/LH ratio was 2.29[1.68,3.12],the E2 level was 22[12,40]pg/ml,the AMH level was 0.06[0.01,0.15]ng/ml,and the AFC was 1[0,2].There were significant statistical differences between the acupuncture group and the acupuncturemedicine combination group in terms of education level,menstrual condition,history of Chinese medicine treatment,history of HRT treatment,FSH level,LH level and AMH level(P<0.05).(2)Intervention situation503 POI patients received a total of 14474 acupuncture treatments,the median duration of acupuncture was 85 days,the median frequency of acupuncture was 7 times/month,the median cumulative number of acupuncture was 20,the median total number of acupuncture was 21,58%(292/503)of the patients received acupuncture for 30 minutes each time,92%(463/503)of the patients were treated with acupoints for regulating menstruation and pregnancy,and 32%(163/503)of the patients were treated with acupuncture and moxibustion.7.4%(37/503)of the patients were combined with ear acupuncture.The frequency of acupuncture and the cumulative number of acupuncture and moxibustion in the acupuncture-medicine combination group were higher than those in the acupuncture and moxibustion group,and there were significant statistical differences(P<0.05).Of the 109 POI patients in the acupuncture-medicine combination group,87(79.8%)used western medicine and 46(42.2%)used Traditional Chinese Medicine.(3)Clinical pregnancy rate,live birth rate and miscarriage rateAmong 503 POI patients,464 needed fertility.The clinical pregnancy rate was 12.28%(57/464),of which the natural pregnancy rate was 5.39%(25/464)and the IVF rate was 6.89%(32/464).The live birth rate was 10.78%(50/464),of which the natural live birth rate was 4.53%(2 1/464)and the IVF live birth rate was 6.25%(29/464).The miscarriage rate was 12.28%(7/57),among which the spontaneous pregnancy miscarriage rate was 7.02%(4/57)and the IVF miscarriage rate was 5.26%(3/57).The clinical pregnancy rate in the acupuncture group was 12.26%(45/367),the live birth rate was 11.17%(41/367)and the miscarriage rate was 8.89%(4/45);the clinical pregnancy rate in the acupuncture and medicine group was 12.37%(12/367),the live birth rate was 9.28%(9/97)and the miscarriage rate was 25%(3/12).Within 60 months of follow-up,there was no significant difference in the clinical pregnancy rate between the acupuncture group and the acupuncture-medicine combination group(Log-rank P=0.94);patients with AFC≥2 had a significantly higher clinical pregnancy rate than those with AFC≤1(Log-rank P=0.01);patients with FSH ≤40 IU/L had a significantly higher clinical pregnancy rate than those with FSH>40IU/L patients(Log-rank P<0.0001).After robust IPTW adjustment,patients with FSH ≤40 IU/L were significantly more likely to have a clinical pregnancy than those with FSH>40 IU/L(HR=3.60,95%CI:[1.68,7.68],P<0.001).Baseline FSH/LH ratio compared to patients with baseline FSH/LH ratio>2.5 Patients with ≤2.5 had a significantly lower clinical likelihood of pregnancy(HR=0.42,95%CI:[0.20,0.87],P=0.019).Compared with those with baseline AFC≥ 2,patients with baseline AFC≤1 had clinical likelihood of pregnancy Sex was significantly reduced(HR=0.36,95%CI:[0.17,0.75],P=0.006).(4)Amenorrhea rateIn the 503 POI patients,the amenorrhea rate decreased from 47%before acupuncture intervention to 16%,a decrease of 31%,and it was reduced to 16%at the 9th month of follow-up,and then maintained until 30 months.In general,there was no significant difference in menstrual recovery rate between the acupuncture group and the acupuncture-medicine combination group(Log-rank P=0.36),after subgroup stratification,it was found that the number of acupuncture less than 24 times,the menstrual recovery rate in the acupuncture group was significantly higher than that in the acupuncture-medicine combination group(Log-rank P=0.038);Acupuncture times more than 24 times,the menstrual recovery rate in the acupuncture-medicine combination group was significantly higher than that in the acupuncture group(Logrank P=0.027).(5)Effective rate of ovarian function-related indicatorsAfter 12 months of follow-up,the proportion of patients with FSH>40 IU/L decreased from 58%to 38%;the proportion of patients with 25 IU/L<FS≤40 IU/L decreased from 28%to 23%;15 IU/L The proportion of patients with L<FSH ≤25 IU/L increased from 14%to 19%;the proportion of patients with 10<FSH≤15 IU/L increased from 0%to 10%;the proportion of patients with FSH≤10 IU/L increased from 0%increased to 9.6%.After robust IPTW adjustment,compared with acupuncture,the FSH recovery rate of the combination of acupuncture and medicine was significantly improved(HR=2.28,95%CI:[1.22,4.25],P=0.01);57%reduction in FSH recovery rate ≤24(HR=0.43,95%CI:[0.22 to 0.84],P=0.013);patients with baseline LH≤17.5 IU/L compared to baseline LH>17.5 IU/L FSH recovery rate was significantly increased in patients with baseline AMH ≤0.05 ng/ml compared with baseline AMH>0.05 ng/ml(HR=3.14,95%CI:[1.33,7.39],P=0.009)82%(HR=0.18,95%CI:[0.07,0.42],P<0.001).After 12 months of follow-up,the proportion of patients with 2<FSH ≤17.5 IU/L increased from 43%to 55%.Acupuncture or the combination of acupuncture and medicine had no significant effect on the FSH/LH ratio,and the normal rate of FSH/LH increased by only 5%6 months after acupuncture intervention and remained stable thereafter.The combination of acupuncture and medicine had a significant increase in E2 within 3 months of followup.In general,the combination of acupuncture and medicine had a higher advantage than acupuncture in increasing E2;after 12 months of follow-up,low estrogen(E2<25 pg/ml)from 55%to 44%.The combination of acupuncture and medicine was superior to acupuncture in increasing AMH levels(AMH in the acupuncture group).The proportion of patients<0.05 ng/ml increased from 33%before intervention to 64%after 12 months of follow-up;acupuncture and medicine combination group,AMH The proportion of patients<0.05 ng/ml decreased from 65%before intervention to 45%after 12 months of follow-up).Acupuncture and acupuncture-medicine combined significantly increased AFC within 3 months of follow-up,and acupuncture-medicine combined had a significantly higher advantage in increasing AFC than acupuncture and moxibustion after 12 months of follow-up(P=0.022).Acupuncture and acupuncture combined showed a significant difference in increasing AFC(Log-rank P=0.005).After stratified analysis according to the number of interventions in subgroups,when the number of acupuncture times≤24 times,the possibility of acupuncture increasing AFC was significantly higher than that of the combination of acupuncture and medicine;It shows that the effect of acupuncture in increasing AFC is reflected in the short-term effect.After robust IPTW adjustment,acupuncture times ≤24 times were significantly more likely to increase AFC compared with acupuncture times>24 times(HR=2.05,95%CI:[1.38,3.02],P<0.001);Patients with a history of childbirth were significantly more likely to have an increased AFC compared with those with a history of childbirth(HR=2.48,95%CI:[1.23 to 5.02],P=0.011);compared with those with a baseline AFC≥ 2,baseline AFC≤One patient was 49%less likely to increase AFC(HR=0.51,95%CI:[0.32 to 0.82],P=0.005).(6)Clinical effectivenessTaking clinical pregnancy as a reference,the effective combination of menstrual recovery and AMH had the highest accuracy(AUC[95%CI]=0.67[0.62,0.72]).Taking the recovery of menstruation and AMH>0.05 ng/ml or clinical pregnancy as the criterion for clinical efficacy,the results showed that the clinical efficacy rate after acupuncture intervention was 51.5%.After robust IPTW adjustment,patients with a history of HRT had a 30%reduction in clinical efficacy compared with no history of HRT(HR=0.70,95%CI:[0.52 to 0.94],P=0.018);compared with baseline LH>17.5 Compared with IU/L,the clinical response rate of patients with baseline LH≤17.5 IU/L was significantly improved(HR=1.70,95%CI:[1.16,2.47],P=0.006);The clinical response rate of patients with effective FSH/LH ratio was significantly improved(HR=6.7,95%CI:[4.81,9.37],P<0.001).3 Construction of a pregnancy prediction model for early-onset ovarian insufficiency treated with acupuncture and moxibustion based on machine learningThe dataset of acupuncture treatment of POI includes 3 prediction targets,68 prediction characteristic variables and 503 observations,and 3 machine learning models are constructed,including pregnancy prediction model,pregnancy time prediction model and efficacy prediction model.(1)The best combination of predicted featuresThrough three feature screening methods,the comprehensive ACC value,AUC value and Fβ value finally formed the best Hospital-Patient-acupuncture-Progression(HPAP).HPAP predicts feature combinations.HPAP-clinical pregnancy(HPAPP)included 15 best predictive feature combinations to construct HPAPP model;HPAPpregnancy time(HPAPT)included 12 best predictive feature combinations to construct HP APT model;HPAP-clinical efficacy(HPAPE)included 18 best predictive feature variables were included to construct the HP APE model.(2)Machine Learning Pregnancy Prediction Model(HPAPP Model)Combining 8 algorithms and 4 model evaluation parameters,it was found that the optimal machine learning algorithm for predicting clinical pregnancy was random forest(AUC=0.75,ACC=0.86,Fp=0.61,Logloss=0.35),so the random forest algorithm was used to build the HPAPP model.The feature importance ranking results of the 15 predictive feature variables showed that the effect of AMH had the greatest contribution to the HPAPP model,followed by the total number of acupuncture and the effectiveness of AFC after acupuncture treatment,and the age of abnormal menstruation,the age of the doctor,the reproductive period,and the length of acupuncture for HPAPP.The model also produced a large contribution.(3)Machine Learning Pregnancy Time Prediction Model(HPAPT Model)In the machine learning model of 6 regression algorithms including 12 predicting characteristic variables,the random forest algorithm has the lowest mean square error,so the random forest algorithm is used to build the HP APT model to predict pregnancy time.In both the training cohort and the validation cohort,the duration of acupuncture had the greatest impact on the model,followed by patient age and reproductive period.(4)Machine learning clinical efficacy prediction model(HPAPE model)Combining 8 algorithms and 4 model evaluation parameters,it was found that the optimal machine learning algorithms for predicting clinical efficacy were LR and MLNN(AUC=0.84,ACC=0.75,Fβ=0.75,Logloss=0.50),due to the interpretation of the LR algorithm Therefore,the LR algorithm is used to construct the HP APE model.The feature importance ranking results of the 18 predictive feature variables showed that the age of abnormal menstruation contributed the most to the HPAPP model,followed by age,disease course,baseline AMH,baseline LH and FSH/LH ratio after acupuncture treatment was effective,in addition,reproductive period,doctor’s practice years,AFC validity and doctor’s age also made a great contribution to the HPPAPE model.Conclusions1 The core related symptom groups of POI are:hot flashes,fatigue,easy fatigue,low mood,sleep disturbance,night sweats,chills,infertility,amenorrhea,oligomenorrhea,vaginal dryness,sexual intercourse discomfort,and low libido.2 Determining cutoff values for ovarian decline based on real-world data from a large sample of POI populations:FSH/LH was 2.5,AMH was 0.05 ng/ml,AFC was 1,LH was 17.5 IU/L,E2 was 25 pg/ml.3 Acupuncture or acupuncture and medicine combined to intervene in POI can improve the clinical pregnancy rate and live birth rate,delay the progress of POI,and reduce the rate of ovarian function decline.4 The objective and reliable criteria for clinical efficacy of "menorrhea-pregnancyreproductive hormones" are clinical pregnancy or menstrual recovery and AMH is effective.The main influencing factors of pregnancy and curative effect of acupuncture on POI were frequency of acupuncture,FSH,FSH/LH ratio,LH,AMH and AFC.Acupuncture for POI benefited from FSH ≤40 IU/L,AMH>0.05 ng/ml and AFC>≥2.5 The developed machine learning pregnancy prediction model,pregnancy time prediction model and clinical efficacy prediction model have high accuracy and discrimination ability and have good clinical practicability and application prospects.Visual machine learning models enable individualized,real-time dynamic prediction of POI patients. |
PDF Full Text Request