Results
We included a total of 325 patients with Chronic rhinosinusitis with nasal polyps (CRSwNP) from June 2018 to July 2020 at the A, B, C hospital. Included patients were following the doctor’s instructions and had a follow up till 1 year. The enrolled patients were randomly assigned to a training (n=195) and validation cohort (n=130). The nomogram was based on the training cohort and its accuracy was internally validated through the validation cohort. The baseline characteristics of the CRS patients between the training cohort and validation cohort are shown in table 1. No significant differences were observed for these characteristics between the training and validation cohort. Univariate analyses were done with the primary objective to confirm the statistical effect between each covariate and the endpoints. Results showed that most covariates had statistically significant associations with the endpoints, except for age, gender, and smoking (Table 2).
Nomogram development
After the initial univariate analyses with extensive review of the medical literature, we included all the covariates in the subsequent multivariate logistic regression models, except for age, gender, smoking, tissue eosinophil counts, preoperative Lund Kennedy score, and Lund Mackay Score. Based on these factors, the nomogram was constructed for calculating the risk of recurrence of the CRS after operation 1year (Figure1A).
A case demonstrating our nomogram usage is shown in Figure1B. For example, if the patient had tissue eosinophil ratio >=10%, low blood eosinophilia, no AR, and asthma, then the total points would be 196 with the corresponding risk of recurrence at 46.11%.
Nomogram validation
Both internal and external validation of the nomogram was performed in this study. The plotted calibration curves correspond to the ideal plot (45°line), which reveals a favorable agreement on the nomogram estimation and the actual observation regarding the probability of uncontrolled disease after the 1 year of post endoscopic sinus surgery. In the training cohort, the nomogram showed the highest accuracy with an AUC of 0.760 (95% CI, 0.688-0.830) (Figure 2.A). The corresponding calibration plot indicates the similarity in the estimation made by the nomogram and clinical findings made during the follow-up period for the recurrence of CRSwNP (Figure 2B). In the validation cohort, the nomogram prediction was 0.635 (95% CI, 0.537-0.733) (Figure 3A). The calibration curve showed a concurrence of predicted probability with the actual probability (Figure 3B).
To assess the clinical applicability of our risk prediction nomogram, clinical impact curve analysis (CICA) and decision curve analysis (DCA) was also performed. The CICA and DCA visually exhibited that the nomogram had superior practical ranges of threshold probabilities and an overall net benefit in terms of outcome for the impacted patient (Figure 4A and 4B).