The study was designed around a deceptively simple question: what happens to patients who arrive at the hospital with an intracranial tumor? In-hospital death is a clinically important short-term outcome because it captures acute neurological deterioration, the timing of treatment decisions, and the real-world capacity of hospital care pathways. In high-income countries, registries and hospital databases routinely track such outcomes, allowing clinicians to benchmark performance and identify modifiable weaknesses. In much of sub-Saharan Africa, however, outcome reporting from neuro-oncology settings has been limited, leaving clinicians and policymakers to operate largely without local evidence. The Luanda team set out to help close that gap using the records their own institution had accumulated over six years.

Methodologically, the researchers treated each admission as the unit of analysis rather than each patient. Because the anonymized institutional registry does not allow reliable patient-level linkage, the number of unique individuals and any repeat admissions could not be determined, a limitation the authors acknowledge explicitly. The primary outcome was death during the hospital stay. Univariable analyses examined age, sex, whether the patient underwent surgery, length of stay, the diagnostic group recorded in the registry, and the year of admission. Reporting followed the STROBE and RECORD guidelines, the international standards for strengthening the reporting of observational studies conducted with routinely collected health data, which lends the analysis a transparency often missing from small single-center studies.

The statistical challenge the team faced was one familiar to anyone working with rare events: only 21 deaths occurred across the entire cohort, fewer than ten events per candidate variable. That number is too small to support a conventional multivariable model with many predictors, which would risk producing unstable, unreliable estimates. Instead, the researchers pre-specified a restricted sensitivity analysis using Firth penalized logistic regression, a technique designed specifically for small-sample and sparse-data settings because it reduces the small-sample bias that plagues ordinary maximum-likelihood estimation. The penalized model was limited to just three predictors: age per ten-year increase, surgery, and a high-risk registry diagnosis category defined as glioblastoma or metastatic disease versus other registry diagnoses.

The results pointed in a consistent direction. Age of 50 years or older was associated with in-hospital death in univariable analysis, with an odds ratio of 4.51 and a 95 percent confidence interval of 1.61 to 12.67, and admissions ending in death were also older when age was analyzed as a continuous variable. Mortality differed across the registry diagnostic groups overall. In the penalized complete-case model, which included 89 admissions and 21 deaths, the high-risk registry diagnosis remained associated with death after adjustment, carrying an odds ratio of 5.70 with a 95 percent confidence interval of 1.58 to 20.52. Age, by contrast, became statistically imprecise in the adjusted model, with an odds ratio of 1.32 per ten years that crossed the null, and surgery showed no significant association, with an odds ratio of 0.79.

The authors are careful, and rightly so, about how far these numbers can be pushed. With only 21 events, the estimates are exploratory and susceptible to residual confounding, and the study does not establish a ranking of determinants of death. The apparent null association for surgery is a particularly instructive example of why crude associations in observational data can mislead. Patients selected for surgery likely had better performance status and more resectable tumors in the first place, so the absence of a statistical link between surgery and mortality should not be interpreted causally. In other words, the healthiest patients were probably the ones offered operations, which would mask any independent effect of the procedure itself.

Sex, surgery, and length of stay were not associated with death in the univariable analyses, a finding that may surprise readers accustomed to high-income datasets where treatment delays and operative variables often dominate outcome discussions. But the Angolan context matters here. Hospital do Prenda functions as a tertiary referral center, and the patients who reach its neurosurgery service represent a selected fraction of everyone in the country with an intracranial tumor. Many patients with suspected brain tumors in Angola may never reach neurosurgical evaluation at all, dying at home or in facilities without neurosurgical capacity. The 22.6 percent mortality figure therefore describes what happens among those who make it through the door, not the full burden of intracranial tumors in the population.

The study’s six-year window, spanning 2020 through 2025, also encompasses the COVID-19 pandemic era, a period when health systems across Africa faced extraordinary strain on hospital beds, staffing, and referral pathways. The researchers examined admission year as a variable but did not identify it as a driver of mortality in their reported findings. What the study does establish is a baseline, a documented starting point against which future improvements in Angolan neuro-oncology care can be measured. The authors note that the registry-based diagnostic groups showed overall differences in mortality, with the high-risk category of glioblastoma and metastatic disease standing out, which aligns with what global neuro-oncology would predict: the most aggressive tumors kill fastest, especially when treatment options are constrained.

Beyond its specific findings, the research carries a broader significance for the field of global neurosurgery. Outcome data from African neuro-oncology settings remain limited, and every carefully documented cohort adds to a picture that international guidelines and resource-allocation decisions currently lack. The Luanda team’s decision to follow STROBE and RECORD reporting principles, to pre-specify a penalized regression strategy appropriate to their small number of events, and to publish with explicit caveats about residual confounding offers a model for how single-center studies in resource-limited settings can meet rigorous methodological standards. The work received no specific external funding, and the authors thank the clinical and administrative teams of Hospital do Prenda for maintaining the institutional records that made the analysis possible.

For clinicians in Luanda and comparable settings, the practical message is that older patients and those with registry diagnoses of glioblastoma or metastatic disease represent the groups at highest risk of dying during admission, and that care pathways, from imaging and biopsy capacity to palliative and neurocritical care, should be oriented accordingly. For researchers, the message is that larger, multi-center, and prospectively designed cohorts are needed to move from hypothesis-generating associations to actionable evidence. The authors themselves frame their findings as exploratory, a scientific foothold rather than a final word. Yet in a field where Angolan data have been nearly absent, even a foothold represents progress, and the 22.6 percent in-hospital mortality figure now stands as a documented reality that future studies, interventions, and health policies in Angola can be measured against.

Subject of Research: In-hospital mortality among intracranial tumor admissions at a tertiary hospital in Luanda, Angola

Article Title: In-hospital mortality among intracranial tumor admissions at a tertiary hospital in Luanda, Angola: a retrospective cohort study

Article References: Mayor, B. S., Xavier, M., Miguel, M. C., Pinto, C., & Neto, S. (2026). In-hospital mortality among intracranial tumor admissions at a tertiary hospital in Luanda, Angola: a retrospective cohort study. BMC Cancer. https://doi.org/10.1186/s12885-026-17039-2

Image Credits: AI Generated

DOI: 10.1186/s12885-026-17039-2

Keywords: intracranial tumors, in-hospital mortality, neuro-oncology, brain cancer, glioblastoma, metastatic disease, neurosurgery, Angola, Africa, retrospective cohort, prognosis, global health