The researchers pooled incidence on the logit scale using random-effects models, combined adjusted risk factors only when at least two independent cohorts used the same exposure definition, and ensured that overlapping cohorts contributed no more than one estimate each. Risk of bias was assessed with QUIPS for prognostic factor studies, PROBAST for prediction models and ROBINS-I for interventional evidence, while the certainty of each finding was graded with GRADE. Heterogeneity was substantial, with an I-squared statistic of 93 percent.

The only moderator that explained part of the variation was study quality: cohorts judged to be at high risk of bias reported an incidence of 11.6 percent, compared with 7.8 percent in cohorts at moderate risk of bias, a difference that was statistically significant at P = 0.02. No other examined moderator accounted for the spread in estimates, which the authors interpret as wide genuine variation between settings and populations rather than a single universal rate. Several patient characteristics and perioperative events emerged as risk factors.

Chronic obstructive pulmonary disease carried the largest association, with an odds ratio of 7.70 (95 percent CI 2.61 to 22.71). Hypoalbuminaemia, defined as a serum albumin below 35 g/L, was associated with an odds ratio of 7.01 (95 percent CI 4.37 to 11.25). Postoperative admission to an intensive care unit was linked to an odds ratio of 2.79 (95 percent CI 2.13 to 3.66), ASA physical status of 3 or higher to an odds ratio of 2.50 (95 percent CI 1.44 to 4.34), and a history of stroke to an odds ratio of 1.92 (95 percent CI 1.34 to 2.76).

Two continuous measures also mattered: each additional year of age raised the odds by a factor of 1.041, and each additional day of delay between fracture and surgery by a factor of 1.091. The certainty of this evidence was modest. GRADE assessments rated the findings for older age and longer fracture-to-surgery delay as moderate certainty, while the associations for COPD, hypoalbuminaemia, ASA status, prior stroke and postoperative ICU admission were rated low certainty.

Because all of these estimates come from observational cohorts, they indicate associations rather than proven causal effects, and residual confounding cannot be excluded. On prevention, the review identified only two before-and-after studies of perioperative prevention protocols. Pooling their results suggested that such protocols were associated with fewer pneumonias, with a risk ratio of 0.60 (95 percent CI 0.44 to 0.82), meaning roughly a 40 percent relative reduction in the settings studied.

However, the certainty of this evidence was rated very low. Before-after designs cannot rule out secular trends, changes in case mix or other concurrent interventions, so the authors conclude that prevention bundles require controlled evaluation before they can be recommended as established practice. Prediction tools fared less well still.

The review included 12 studies of models designed to predict which patients would develop postoperative pneumonia, and all 12 were judged to be at high risk of bias using the PROBAST instrument. Only three of the models had undergone external validation in an independent population. Because the authors deliberately synthesised prediction-model performance without pooling areas under the curve, no combined accuracy figure is reported; the key takeaway is that the available models are not ready for routine clinical use.

The clinical message is nonetheless actionable in a limited sense. The identified factors, particularly COPD, hypoalbuminaemia, ASA physical status of 3 or above, prior stroke and longer surgical delay, could help clinicians identify patients who might benefit from intensified preventive measures after hip fracture surgery. The authors are careful to frame this as targeting of prevention based on observational associations of low to moderate certainty, not as a validated risk score or a proven treatment pathway.

The study's scope and limitations are clearly delineated. It covers older adults undergoing hip fracture surgery, draws on cohorts published in a defined five-and-a-half-year window, and its incidence estimate is shaped by the quality of the underlying studies, as shown by the difference between high- and moderate-risk-of-bias cohorts. The very low certainty attached to the prevention evidence and the absence of externally validated, low-bias prediction models mean that the review maps the problem and its correlates rather than delivering ready-made solutions.

Its authors call for controlled trials of prevention bundles and better-validated prediction models, positioning the current findings as a baseline against which future interventional and modelling research can be measured.

Subject of Research: Postoperative pneumonia after hip fracture surgery in older adults: incidence, risk factors, prevention and prediction models

Article Title: Incidence, risk factors, prevention and prediction of postoperative pneumonia after hip fracture surgery in older adults: a systematic review and meta-analysis

Article References: He, H., Zuo, Z., & Chen, X. (2026). Incidence, risk factors, prevention and prediction of postoperative pneumonia after hip fracture surgery in older adults: a systematic review and meta-analysis. BMC Geriatrics. https://doi.org/10.1186/s12877-026-08543-y

Image Credits: AI Generated

DOI: 10.1186/s12877-026-08543-y

Keywords: hip fracture, postoperative pneumonia, older adults, meta-analysis, systematic review, risk factors, COPD, hypoalbuminaemia, prediction models, prevention bundles, geriatric surgery, GRADE certainty