Minimal Gain in Prognostic Certainty for Patients With Hematologic Malignancies Over 2 Weeks of Intensive Care: An...

We aimed to examine how the predicted 1-year survival and its prognostic certainty evolve during the first two weeks of ICU admission.

Retrospective cohort study.

Two academic medical centers in The Netherlands. External validation in the Medical Information Mart for Intensive Care-IV database from a tertiary medical center in the United States.

Patients with active hematologic malignancies admitted to the ICU.

None.

Separate prediction models for 1-year survival were developed using data available at day 1, 7, and 14 after ICU admission for 555, 181, and 94 ICU admissions resulting in an area under the receiver operating characteristics curves of 0.71, 0.67, and 0.66, respectively. At the individual patient level, prognostic certainty quantified by entropy increased meaningfully (entropy decrease > 0.25) in 2% of patients between day 1 and day 7 (in an additional 12% certainty increased because of death) and in 14% of patients between day 7 and day 14 (in an additional 18% because of death). Among patients alive on day 1, 2% of patients with an "uncertain" and 10% with a "poor" initial prognosis had shifted to a more favorable category by day 7. Of the patients alive and still in the ICU on day 7, 31% of patients with an "uncertain" and 16% with a "poor" prognosis had shifted to a more favorable category by day 14. Results in the external validation cohort were comparable.

In patients with hematologic malignancies admitted to the ICU, prognostic certainty about long-term survival increased little during in the first 2 weeks of ICU admission, aside from increases in prognostic certainty due to early mortality. Despite the use of rich ICU datasets and different state-of-the-art modeling strategies, overall model performance was modest, suggesting that prognosis in this population is largely driven by disease-related and patient-specific factors beyond the ICU course.