Applying a risk prediction model for bloodstream infection in a febrile, nonseverely neutropenic cohort of pediatric stem cell transplant patients.

BACKGROUND

The optimal management of febrile stem cell transplant (SCT) patients presenting without severe neutropenia (absolute neutrophil count [ANC] ≥ 500/µL) is unclear. The authors have developed iterative risk prediction models (Esbenshade Vanderbilt [EsVan] models) that reliably predict bloodstream infections (BSIs) in the febrile general pediatric oncology population without severe neutropenia, but SCT-specific data are limited.

METHODS

All SCTs occurring from May 2005 to November 2019 at a single institution were identified. Episodes of fever with a central venous catheter and ANC values ≥ 500/µL were abstracted. All previous versions of the EsVan model were applied to the SCT data, and c-statistics were generated. The models were additionally applied to each type of transplant (autologous/allogeneic), and a new allogeneic model that further adjusted for metrics of immunosuppression, Esbenshade Vanderbilt Allogeneic SCT Model (EsVanAlloSCT), was developed and internally validated.

RESULTS

For 429 SCT episodes (221 autologous and 208 allogeneic), the BSI incidence was 19.6% (84 of 429), and it was higher in allogeneic transplant patients (25.5%) than autologous transplant patients (14.0%; p < .01). All versions of the EsVan model performed well for the overall SCT cohort (c-statistics, 0.759-0.795). The EsVan models performed better for the autologous episodes (c-statistics, 0.869-0.881) than the allogeneic SCT episodes (c-statistics, 0.678-0.717). The new allogeneic transplant-specific model, EsVanAlloSCT, which added an adjustment for the extent of immunosuppression, yielded a c-statistic of 0.792 (bootstrap-corrected, 0.750).

CONCLUSIONS

The EsVan models work exceptionally well when they are applied to autologous SCT, but they work less well for allogeneic SCT. EsVanAlloSCT appears to improve the predictive ability in allogeneic SCT, but it will need additional external validation.