Models A and B were externally validated using 57 chylothorax and 114 non-chylothorax patients from Lucille Packard Children’s Hospital. Compared to the development cohort, the external validation cohort had higher CTO on PCD 1 (p < 0.001), were older (p = 0.026), and were more non-Caucasian (p < 0.001) (Supplementary Table 2). The median days after surgery to diagnosis using pleural fluid testing was 9.7 days (IQR 4.7, 15.3).
External validation of Models A and B was good with c-statistics of 0.84 (95% CI: 0.78, 0.90) and 0.89 (95% CI: 0.84, 0.93) respectively. The Hosmer Lemeshow observed to expected goodness of fit plot shows good model calibration (p = 0.36 and p = 0.71 respectively) (Supplementary Fig. 2).
We report on novel and parsimonious predictive models for chylothorax in pediatric cardiac patients on the PCD 1. The models have good discrimination (c-statistic 0.78 and 0.84) with similar performance when externally validated (c-statistic 0.84). Our models demonstrate that volume of CTO on PCD 1 is an important variable in predicting the odds of chylothorax. The models we developed do not rely on delivery of enteral nutrition, which is a distinct advantage over using pleural fluid testing which is the current diagnostic standard.
Diagnostic levels of chylomicrons and triglycerides are dependent on a regular fat diet due to long-chain triglycerides being directly absorbed from the gut into the lymphatic system.[7, 8, 14] Some centers routinely diagnose chylothorax by clinical appearance of milky fluid. However, about 25% of chylothorax cases do not have the characteristic appearance, especially in fasting patients or those with genetic syndromes or abnormal lymphatic development.[7, 13, 14, 27] Illness severity is directly related to the likelihood that enteral nutrition is withheld. Thus, the most critically ill chylothorax patients may be at highest risk for delayed diagnosis when solely relying on traditional pleural fluid testing or qualitative assessment.
Patient surgical complexity and acuity are directly related to both chylothorax prevalence and associated morbidity.[14, 28, 29] Early chylothorax diagnosis in the postoperative period leads to early resolution, and conservative therapy such as afterload reduction, diuresis, and other measures to prevent venous hypertension or lymphatic obstruction may be more effective when used without delay. [1, 5, 8] For the sickest patients who do not receive enteral nutrition, our models could facilitate detection as early as the first day after sternal closure with more expeditious management when compared to diagnosis that occurred on the third and tenth days after sternal closure in the internal and external cohorts, respectively.
Similar to prior multivariate analysis, we also found a higher prevalence of chylothorax in patients with single ventricle and higher complexity surgeries, longer cardiopulmonary bypass (CPB) times, and younger surgical ages. [5, 23, 28]. However, these prior models did not include any measure of CTO volume. The c-statistic improves from 0.72 to 0.84 when CTO on PCD 1 as a continuous variable is added to the model, demonstrating the importance of this novel covariate in predicting the odds of chylothorax.
We recognize that applying this model could result in some patients being unnecessarily treated for chylothorax. However, since incremental CTO increases chylothorax risk, we would use high CTO to stratify risk and guide confirmatory pleural fluid testing. The potential benefit of mitigating the adverse effects of chylothorax through earlier treatment and resolution would be weighed against the relatively low risk of noninvasive therapies such as fat-modified diet, diuretics, or milrinone, particularly if those therapies are short-lived. It is also possible that the hemodynamic significance of high chest tube output may be underappreciated, and that treatments to alleviate lymphatic or venous hypertension may be beneficial or even preclude chylothorax diagnosis. This analysis shows the potential value of a point-of-care predictive metric for chylothorax in pediatric cardiac intensive care units.
The literature suggests that protocolized management of chylothorax reduces time to diagnosis, treatment duration and time without feeds, chest tube and central line utilization, duration of mechanical ventilation and length of ICU and hospital stay.[1, 3, 30, 31] Incorporation of our model in protocols to trigger pleural fluid testing irrespective of milky appearing pleural fluid could allow for earlier detection and potentially earlier treatment and resolution. We look to incorporate inflammatory and immune mediated biomarkers into future iterations, investigate the influence of physiologic risk profiles including fluid balance, central venous pressures and postoperative pulmonary hypertension on model discrimination.
Limitations include those inherent to retrospective analyses, which were used for both model development and external validation. Additionally, there is potentially a confounder effect due to unknown covariates with incompletely understood pathophysiology of chylothorax and covariates reported in other studies that we did not collect such as thrombosis data, central line use, hemodynamic and nutritional data, or specific cardiac lesions. Given the variable incidence of chylothorax across centers it is also plausible that the etiology may also vary and thus high CTO may not be an early manifestation at other centers.[1, 3, 4, 7, 8, 11, 14, 21, 28, 32, 33] However, our external validation results suggest the model is generalizable.
High volume chest tube output on the first day after sternal closure may predict chylothorax earlier than traditional diagnosis by pleural fluid testing, independent of enteral nutrition. Early detection and treatment can potentially decrease time to resolution, which may have the greatest impact on the most critically ill patients who are not fed enterally.