Development and Comparison of Predictive Models Based on Different Types of Influencing Factors to Select the Best One for the Prediction of OSAHS Prevalence

Abstract

Objective: This study aims to retrospectively analyze numerous related clinical data to identify three types of potential influencing factors of obstructive sleep apnea-hypopnea syndrome (OSAHS) for establishing three predictive nomograms, respectively. The best performing one was screened to guide further clinical decision-making.

Methods: Correlation, difference and univariate logistic regression analysis were used to identify the influencing factors of OSAHS. Then these factors are divided into three different types according to the characteristics of the data. Lasso regression was used to filter out three types of factors to construct three nomograms, respectively. Compare the performance of the three nomograms evaluated by C-index, ROC curve and Decision Curve Analysis to select the best one. Two queues were obtained by randomly splitting the whole queue, and similar methods are used to verify the performance of the best nomogram.

Results: In total, 8 influencing factors of OSAHS have been identified and divided into three types. Lasso regression finally determined 6, 3 and 4 factors to construct mixed factors nomogram (MFN), baseline factors nomogram (BAFN) and blood factors nomogram (BLFN), respectively. MFN performed best among the three and also performed well in multiple queues.

Conclusion: Compared with BAFN and BLFN constructed by single-type factors, MFN constructed by six mixed-type factors shows better performance in predicting the risk of OSAHS.

Keywords: blood glucose; blood lipid; nomogram; obstructive sleep apnea-hypopnea syndrome; risk prediction.

FIGURE 1

The flow chart of the entire study.

FIGURE 2

Correlation and difference analysis of 11 clinical continuous indicators. (A) Correlation analysis between AHI and 11 continuous indicators. Different sizes and colors of circles represent different correlation coefficients and significance p-value, respectively. (B) Differences of 11 continuous indicators between OSAHS and non-OSAHS groups. Different symbols are shown at the top of the block diagram. ns: p > 0.05; *p < 0.05; **p < 0.01; ***p < 0.001.

FIGURE 3

The forest plot shows the univariate logistic regression results of 16 indicators. The red and blue boxes represent the corresponding Odds ratio greater than 1 and less than 1, respectively.

FIGURE 4

Different types of influencing factors (6 mixed factors, three baseline factors, four blood factors) selection process using lasso regression for nomogram, respectively. (A,B) Six mixed factors. (C,D) Three baseline factors. (E,F) Four blood factors. (A,C,E) The optimal parameters (lambda) in 3 LASSO models are selected with 10-fold cross-validation via minimum criteria, respectively (59). Draw vertical dashed lines at the optimal value using the minimum criteria and the 1 SE of the minimum criteria (47). (B,D,F) Distribution diagrams of LASSO coefficients for three types of factors generated by log (lambda) sequence, respectively (47). Vertical lines are drawn at the best lambda to select 6, 3, and 4 factors with non-zero coefficients, respectively.

FIGURE 5

Three types of factors screened by lasso regression were used to construct MFN, BAFN and BLFN, respectively. (A) MFN. (B) BAFN. (C) BLFN.

FIGURE 6

Comparison of the prediction performance of MFN, BAFN and BLFN. (A) Calibration curves of mixed results. The x-axis and y-axis represent the risk predicted by the nomogram and the risk diagnosed. The diagonal dashed line represents the perfect prediction of the ideal model (47). The black, yellow and blue solid lines represent the performance of MFN, BAFN and BLFN, respectively, of which a closer fit to the diagonal dotted line represents a better prediction. (B) ROC curves of mixed results. The X axis and Y axis represent 1-specificity and sensitivity, respectively. The lower right corner of the figure shows the corresponding AUC values of MFN, BAFN and BLFN, and the 95% confidence interval, respectively. (C) DCA curves of mixed results. The y-axis measures the net benefit (47). The thin solid gray line and the thin black solid line represent the assumptions that all patients rely on and do not rely on MFN predictions, respectively. The black, yellow, and blue thick solid lines represent the predictions of MFN, BAFN, and BLFN, respectively.

FIGURE 7

The performance test of MFN prediction in the training, test and the whole queues, respectively. (A–C) Calibration curves. The black solid line represents the performance of the nomogram. (D–F) ROC curves. The X axis and Y axis represent false positive and true positive rates, respectively. The top of the figure shows the corresponding AUC value. (G–I) DCA curves. The blue thick solid lines represent the predictions of MFN.