Predictive nomogram for early detection of invasive fungal disease deterioration --- a 10-year retrospective cohort study

Abstract

Background: Invasive fungal disease (IFD) is characterized by its capacity to rapidly escalate to life-threatening conditions, even when patients are hospitalized. However, the precise prognostic significance of baseline clinical characteristics related to the progression outcome of IFD remains elusive.

Methods: A retrospective cohort study spanning a duration of 10 years was conducted at two prominent tertiary teaching hospitals in Southern China. Patients with proven IFD were queried and divided into serious and non-serious groups based on the disease deterioration. To establish robust predictive models, patients from the first hospital were randomly assigned to either a training set or an internal validation set, while patients from the second hospital constituted an external test set. To analyze the potential predictors of IFD deterioration and identify independent predictors, the study employed the least absolute shrinkage and selection operator (LASSO) method in conjunction with binary logistic regressions. Based on the outcomes of this analysis, a predictive nomogram was constructed. The performance of the developed model was thoroughly evaluated using the training set, internal validation set, and external test set.

Results: A total of 480 cases from the first hospital and 256 cases from the second hospital were included in the study. Among the 480 patients, 81 cases (16.9%) experienced deterioration, and out of those, 45 (55.6%) cases resulted in mortality. Seven independent predictors were identified and utilized to construct a predictive nomogram. The nomogram exhibited excellent predictive performance in all three sets: the training set, internal validation set, and external test set. The area under the receiver operating characteristic curve (AUC) for the training set was 0.88, for the internal validation set was 0.91, and for the external test set was 0.90. The Hosmer-Lemeshow test and Brier score indicated a high goodness of fit for the model. Furthermore, the calibration curve demonstrated a strong agreement between the predicted outcomes from the nomogram and the actual observations. Additionally, the decision curve analysis exhibited that the nomogram provided significant clinical net benefits in predicting IFD deterioration.

Conclusions: The study successfully identified seven independent predictors and developed a predictive nomogram for early assessment of the likelihood of IFD deterioration.

Keywords: Independent predictor; Invasive fungal disease; Nomogram; Prediction probability.

Fig. 1

Patient enrolment and study flowchart. Notes: IFD, invasive fungal disease; LASSO, least absolute shrinkage and selection operator

Fig. 2

Selection of optimal predictors using the LASSO regression. Notes: a Determination of optimal regularization parameter λ of the LASSO model to select the optimal predictors for IFD deterioration. When log(λ) is -3.79, LASSO shows the best predictive performance with the minimum number of predictors and regression errors, where the coefficients of 27 out of 43 baseline variables are shrunk to zero. b The coefficient profile of LASSO. The trajectory of each IFD-related feature coefficient is observed in the LASSO coefficient profiles with the changing of the λ values. LASSO, least absolute shrinkage and selection operator; IFD, invasive fungal disease

Fig. 3

Predictive nomogram for early detection of IFD deterioration. Notes: Each independent predictor is represented as 0 for "no" and 1 for "yes", and the corresponding score is marked on the uppermost points axis. Total score for all predictors can be obtained by adding the scores of each predictor, and the bottom horizontal axis shows the corresponding risk of IFD deterioration. IFD, invasive fungal disease; SOT, solid organ transplant; CHD, chronic heart disease; Comorbidities, one or more comorbidities

Fig. 4

Receiver operating characteristic curves and calibration curves of the nomogram. Notes: Figure a, b, and c indicate the AUCs of the nomogram, with sensitivities and specificities, and d and e and f denote the calibration curves in the training set, internal validation set and external test set, respectively. The horizontal axis of figure d, e, and f show the predicted probability of IFD deterioration and the vertical axis show the actual probability. The predictive performance of the nomogram (black line) closer to the ideal prediction line (dotted line) represents a higher predictive accuracy. AUC, area under the receiver operating characteristic curve

Fig. 5

Decision curve analysis of the prediction nomogram. Notes: Decision curve analysis of the nomogram prediction in the training set (a), internal validation set (b), and external test set (c). The vertical axis represents the net benefit, while the horizontal axis denotes the threshold probability. The black horizontal line corresponds to the “treat none” strategy, assuming no patients receive intervention. The gray slanted line corresponds to the “treat all” strategy, assuming all patients are predicted to be positive and thus receive intervention. The farther the decision curve is above these two reference lines, the greater the net clinical benefit of the model. IFD, invasive fungal disease; SOT, solid organ transplant; CHD, chronic heart disease; Comorbidities, one or more comorbidities