Development and validation of a new nomogram for self-reported OA based on machine learning: a cross-sectional study

Abstract

Developing a new diagnostic prediction model for osteoarthritis (OA) to assess the likelihood of individuals developing OA is crucial for the timely identification of potential populations of OA. This allows for further diagnosis and intervention, which is significant for improving patient prognosis. Based on the NHANES for the periods of 2011-2012, 2013-2014, and 2015-2016, the study involved 11,366 participants, of whom 1,434 reported a diagnosis of OA. LASSO regression, XGBoost algorithm, and RF algorithm were used to identify significant indicators, and a OA prediction nomogram was developed. The nomogram was evaluated by measuring the AUC, calibration curve, and DCA curve of training and validation sets. In this study, we identified 5 predictors from 19 variables, including age, gender, hypertension, BMI and caffeine intake, and developed an OA nomogram. In both the training and validation cohorts, the OA nomogram exhibited good diagnostic predictive performance (with AUCs of 0.804 and 0.814, respectively), good consistency and stability in calibration curve and high net benefit in DCA. The nomogram based on 5 variables demonstrates a high accuracy in predicting the diagnosis of OA, indicating that it is a convenient tool for clinicians to identify potential populations of OA.

Keywords: Machine learning; NHANES; Nomogram; Osteoarthritis.

Fig. 1

Flow chart of sample selection.

Fig. 2

Selection of main predictors of OA. (A) Selection of the tuning parameter lambda in the LASSO regression via 10-fold cross-validation based on minimum criteria. Misclassification error from the LASSO regression cross-validation procedure was plotted as a function of log lambda. The y-axis indicates the misclassification error. The x-axis indicates the log lambda. (B) The LASSO coefficient profiles of clinical features. The dotted vertical line was plotted at the value selected using 10-fold cross-validation in A. The resulting variables with non-zero coefficients are indicated in the plot. (C) Importance ranking of 7 variables by LASSO regression. (D) Importance ranking of 15 variables via 10-fold cross-validation and assessment of feature importance by XGBoost algorithm. (E) Selection of variables by RF algorithm. Mean decrease accuracy and mean decrease gini from the RF algorithm cross-validation procedure was plotted. (F) Importance ranking of 15 variables via assessment of feature importance by RF algorithm. (G) Intersection of variables obtained from the three algorithms, and come out with 5 main predictors of OA.

Fig. 3

The nomogram represents the predicted probability of OA on a scale of 0 to 200. For each predictor, draw a vertical line straight up to the point axis and note the corresponding points. Sum the points from each predictor, and the total score corresponding to a predicted probability of OA can be found at the bottom of the nomogram.

Fig. 4

The performance of the new nomogram for predicting OA. A ROC curve in the training cohort. The x-axis is 1-Specifcity; the y-axis is the Sensitivity. B Calibration curve in the training cohort. The x-axis is the nomogram predicted probability of OA; the y-axis is actual probability. C ROC curve in the validation cohort. D Calibration curve in the validation cohort.

Fig. 5

The clinical utility of the nomogram was evaluated by DCA. A Decision curve in the training cohort. B Decision curve in the validation cohort. The x-axis represents the threshold probability. The y-axis represents net benefits.