Material and Methods
This is a retrospective and nonconcurrent cohort study. The study was approved by the local Ethics Committee ([2017]164). According to the European Position Paper on Rhinosinusitis and Nasal Polyps 2012 (EPOS2012) guidelines, patients who satisfied the diagnostic criteria of chronic rhinosinusitis with nasal polyps were included in the study fromA, B, C hospital. All patients received FESS between January 2018 and December 2020 and were periodically reassessed during their routine outpatient visits following the surgery. These patients were initially treated with AMT i.e., nasal steroids (drops/sprays/rinses), saline rinses, educated regarding technique, oral corticosteroid short-course (OCS), and two-course antibiotics before surgery.
Patients were instructed to use topical corticosteroids-budesonide nasal spray (256ug/day for 6 months), and intranasal budesonide suspension (1mg/day for 4 weeks) after surgery. They were reassessed periodically at their routine outpatient visits at 1 to 3 months after surgery then once in 3 months until 1 year follows up. During the assessment in the follow-up visits, if their symptoms or endoscopic signs persisted, they received new AMT i.e., nasal steroids (drops/spray/rinses), saline rinses, education regarding technique, OCS, and optional two-course antibiotics. The symptoms, endoscopic scores, and modified treatment (if any) were recorded by clinicians after 1 year.
Items recorded from the enrolled patients were as following:
Data collection
Patients were divided into 2 groups of controlled (included partly controlled) and uncontrolled CRS, based on the disease control criteria of EPOS2020. Patients were followed up for 1 year after surgery, until the end of the study period (30th December 2020). Time-to-event was defined as the time starting from surgery till the 12th month post-operatively. According to the EPOS2020, the control criteria of the CRS can be divided into symptoms, nasal endoscopy, the need for recuse treatment. Symptom substituted by ‘VAS (Visual Analogue Scale)< 5’, and ‘present/impaired’ by ‘VAS ≥ 5. Furthermore, the detailed symptoms related to CRS are included in supplement table S1. The evaluation endpoint was 12th month post-operatively.
Nomogram development
The nomogram model was formulated by the results of multivariate analysis. Univariate analysis with a significant difference at P-value(<0.05) between all variables was included in the multivariate analysis. The P-value <0.05 in multivariate analysis was also included as the prognostic factor in the nomogram. AR and PBEC were statistically significant in univariate analysis for 1 year disease control but not significant difference in multivariate analysis for 1 year diease control. However, AR and PBEC have long been recognized to determine the prognosis of CRS. AR and PBEC were also included in the nomogram for the current study, since excluding these covariates would have over-inflated the effects of the remaining factors and decrease the predictive power of our model. The Cox proportional hazard model was used to produce nomograms for predicting the risk of the uncontrolled incident after the surgery. A score based on regression coefficients was assigned to these factors.
Model evaluation
The nomogram’s forecast performance was evaluated by the receiver operating characteristic (ROC), the area under curve (AUC) for both training and validation cohort. In a logistic regression model, the value of AUC is the same as that yielded by the concordance index (c-index), with values ranging from 0.5 (no predictive value) to 1.0 (complete discrimination). A larger AUC value represents a more accurate prediction of the uncontrolled disease possibility. The agreement between the predicted uncontrolled incident and the observed uncontrolled incident after bias correction was quantified by the calibration curves of the nomogram for determining the uncontrolled incident rate. Decision curve analysis (DCA) was also carried out to compare the potential net benefit of the predictive models.
Statistical analysis
We compared the patient pathologic characteristics and demographic profile between training and validation cohort by using Fisher’s exact tests and chi-squared tests. Multivariate logistic regression analysis was used to distinguish the independent risk factors associated with uncontrolled disease. Nomogram development was carried out by using the library “rms” in R for MACOS. All statistical analyses were conducted by the R software version 4.0.2 (R Foundation for Statistical Computing, Vienna, Austria; www.R-project.org). The p values<0.05 were considered statistically significant.