Objective To construct two Nomograph risk factor predictive models for chronic heart failure patients with renal insufficiency based on Lasso-Logistic regression analysis and compare their efficacy.

Methods The clinical data of 996 patients with chronic heart failure were collected. These patients were randomly divided into modeling group(698 cases) and verification group(298 cases) in a ratio of 7∶3. Lasso regression was used to screen variables, multivariate Logistic regression was used to screen independent risk factors for variables with statistical significance, and two models were compared to the evaluate their clinical effectiveness.

Results Of 698 patients in the modeling group, 148(21.20%) were complicated with renal insufficiency. Multivariate Logistic regression results of model 1 showed that hemoglobin, creatinine, uric acid, age, valvular heart disease, and presence or absence of complication were independent risk influencing factors(P < 0.05). Multivariate Logistic regression results of model 2 showed that hemoglobin, creatinine, uric acid, and presence or absence of complication were independent influencing factors (P < 0.05). The area under the curve (AUC) of model 1 was 0.814, and Hosmer-Leishow test results showed that it did not deviate, and was perfectly matched (P=0.08), the calibration chart showed that the model has good consistency. The AUC of model 2 was 0.806. The results of Hosmer-Lemeshow showed that the model was deviated from the perfect fit (P < 0.01), and the calibration chart showed that the consistency of the model was poor. The results of the validation group showed that the AUCs of model 1 and model 2 were 0.835, 0.824, respectively. Hosmer Lemeshow test showed that the models did not deviate from the perfect fit (P=0.12, 0.45), and the calibration curve also had good consistency.

Conclusion The two Nomograph risk factors predictive models based on Lasso-Logistic regression have better ability in predicting the risk of chronic heart failure patients with renal insufficiency, but model 1 has better differentiation, consistency beween Hosmer-Lemeshow test results and calibration curve, stronger clinical applicability and higher net benefit, so model 1 is recommended for clinical application.