When oncologists at a Japanese acute care hospital documented how long they expected a patient with advanced cancer to survive, their estimates alone missed a large share of deaths that occurred within weeks. A new retrospective study from the University of Occupational and Environmental Health Hospital shows that adding the Palliative Prognostic Index, a bedside score built from performance status, oral intake, edema, dyspnea at rest, and delirium, caught many of these overlooked cases, although the benefit came with a clear cost in specificity.
The study, published in Cancer Reports, included 357 patients with advanced cancer for whom discontinuation of anticancer treatment and transition to best supportive care had already been decided. All were referred to the hospital’s palliative care team between April 2020 and November 2025. The researchers used the date of the documented clinician prediction of survival as the index date and evaluated whether each patient died within 21 or 42 days of that point. The median actual survival from the index date was 26 days, and 44.3 percent of patients died within three weeks while 66.4 percent died within six weeks.
Clinician prediction of survival, or CPS, is the survival estimate written by the attending physician in the medical record. It is not a pure intuition; it draws on cancer type, disease stage, molecular profile, treatment lines and response, and population-level survival data from trials and real-world studies. Yet previous research has repeatedly shown that physicians’ predictions in terminally ill patients tend to overestimate survival, and a pooled analysis found that oncologists’ estimates were reasonably well calibrated for populations but had limited accuracy for individual patients. The practical question in acute care is therefore not whether CPS is accurate in the abstract, but how to supplement it so that short-term mortality risk is recognized early enough to shape care planning, discharge coordination, and transition to palliative care.
The Palliative Prognostic Index was developed by Morita and colleagues as a simple tool for predicting three- and six-week survival in terminally ill cancer patients. It scores five variables that reflect functional decline and symptom burden near the end of life. In this study, the PPI was calculated retrospectively from medical records at the same index date as the CPS, using structured nursing observations, physician notes, and palliative care team documentation. Thresholds of PPI at or above 6.5 and at or above 4.5 were used to flag patients at risk of death within three and six weeks respectively, mirroring the original scoring system.
For deaths within 21 days, CPS alone had a sensitivity of 37.3 percent and a specificity of 89.4 percent, meaning it was reliable when it flagged a patient but missed most of those who died. PPI alone showed the opposite profile, with sensitivity of 58.9 percent and specificity of 72.4 percent. When the researchers classified a patient as positive if either measure crossed its threshold, sensitivity rose to 72.2 percent while specificity fell to 66.3 percent. The pattern repeated at 42 days: CPS sensitivity of 48.9 percent and specificity of 78.3 percent, PPI sensitivity of 73.4 percent and specificity of 58.3 percent, and a combined rule that reached 82.3 percent sensitivity at the price of 47.5 percent specificity.
Receiver operating characteristic analysis told a subtler story. The area under the curve for 21-day mortality was 0.742 for PPI and 0.732 for CPS, and for 42-day mortality the corresponding values were 0.731 and 0.727, with no statistically significant difference between the two tools at either horizon. In other words, the two measures discriminated between survivors and nonsurvivors about equally well overall, but their operating characteristics differed in ways that mattered clinically. CPS was the more specific instrument, PPI the more sensitive one, and the either-positive rule was designed explicitly as a sensitivity-oriented screen rather than a more accurate predictive model.
The four-group classification by CPS and PPI status made the complementary value concrete. For the three-week classification, 21-day mortality was 80.9 percent among patients positive on both measures, 63.6 percent among those positive on CPS only, 54.5 percent among those positive on PPI only, and 25.0 percent among those negative on both. Median survival followed the same gradient, at 8, 16, 18, and 42.5 days respectively. Among the 99 patients who died within 21 days but had been classified as negative by CPS, PPI flagged 55, or 55.6 percent. For the six-week classification, the corresponding figures were 88.0, 61.8, 68.1, and 42.4 percent mortality across the four groups, and PPI identified 79 of 121 CPS-negative patients who died within 42 days, or 65.3 percent.
Kaplan-Meier analyses reinforced the same signal. Among patients classified as negative by CPS, those who were PPI-positive had significantly shorter survival than those who were PPI-negative, with median survival of 18 versus 42.5 days for the three-week classification and 26.5 versus 50 days for the six-week classification. An exploratory analysis of in-hospital deaths pointed in the same direction: of 113 patients who died during hospitalization, 69 had been predicted by CPS to survive more than three weeks, and 40 of those 69 met the PPI threshold for three-week mortality; 32 of those 40 died within 21 days of the index date.
The authors are careful about what these numbers do and do not mean. The combined rule does not provide an individualized probability of death, weight CPS and PPI differently, or account for interactions among prognostic factors, and validated models such as the Palliative Prognostic Score and the Prognosis in Palliative care Study models remain alternatives when the required clinical or laboratory information is available. Most importantly, the double-negative classification carried substantial residual risk: 25.0 percent of patients negative on both measures in the three-week classification and 42.4 percent in the six-week classification still died within the corresponding time horizon. The combined assessment therefore cannot reliably rule out short-term mortality and should not be used to defer care planning.
Several limitations frame the findings. This was a single-center retrospective study restricted to patients referred to the palliative care team, and only those with a confirmed date of death and a quantifiable CPS entered the primary cohort; included patients were younger than excluded patients and differed in treating department and clinical progression pattern, suggesting potential selection bias. PPI was calculated retrospectively from records rather than assessed prospectively, formal inter-rater reliability testing was not performed, and converting free-text prognostic expressions into numerical days may have introduced measurement error near the 21- and 42-day thresholds. The exploratory subgroup analyses involved multiple comparisons without adjustment for multiplicity, and the population, which included some patients who survived a year or longer, was broader than the terminally ill population in which PPI was originally validated. The study also did not test whether CPS/PPI-based reassessment actually improves care planning, discharge coordination, or patient-centered outcomes; a large cluster-randomized trial of structured advance care planning increased specialist palliative care use and documentation but did not improve quality of life. Within those limits, the authors conclude that PPI may serve as a practical adjunct to CPS, a prompt to revisit the estimated prognosis and the urgency of care planning when a patient flagged as low risk by the clinician’s own estimate is, by measurable bedside signs, deteriorating faster than expected.
Subject of Research: Combining clinician survival predictions with the Palliative Prognostic Index to screen for short-term mortality in advanced cancer patients transitioning to best supportive care
Article Title: Combined Use of Clinician Prediction of Survival and the Palliative Prognostic Index for Short‐Term Mortality Screening in Advanced Cancer Patients in an Acute Care Hospital
Article References: Shiraishi, T., Kimura, M., Nabeshima, N., Ide, A., Kondo, T., Hamada, K., Akiyama, M., & Terada, T. (2026). Combined Use of Clinician Prediction of Survival and the Palliative Prognostic Index for Short‐Term Mortality Screening in Advanced Cancer Patients in an Acute Care Hospital. Cancer Reports, 9(11), Article e70742. https://doi.org/10.1002/cnr2.70742
Image Credits: AI Generated
DOI: 10.1002/cnr2.70742
Keywords: palliative care, prognostication, advanced cancer, clinician prediction of survival, Palliative Prognostic Index, best supportive care, acute care hospital, mortality screening, retrospective study, care planning, Combined, Clinician
News Source: Nathaniel Bowman. (November 5, 2026). Simple bedside score flags hidden short-term death risk that doctors miss in advanced cancer. Scienmag.
