Highlight
A Veterans Affairs cohort study derived a preoperative risk calculator that predicts whether patients will experience routine recovery within 60 days or face protracted recovery and long-term loss of independence lasting 6 months or more.
The strongest predictors were the Care Assessment Needs score for 90-day hospitalization and baseline functional status, underscoring the central importance of frailty-related risk and preoperative vulnerability.
Model discrimination was strong, with c-statistics of 0.906 to 0.908 in training and test samples, and both sensitivity and specificity were approximately 83%.
A more complex gradient boosting machine model produced only marginally higher discrimination, suggesting that an interpretable regression-based tool may be sufficient for point-of-care use.
Background
Surgical outcomes research has traditionally emphasized 30-day mortality, complications, readmissions, and length of stay. These measures remain important, but they incompletely capture what many patients value most: whether they return to baseline function, remain independent, and avoid a prolonged period of disability after surgery. This issue is particularly relevant for older adults and medically complex patients, in whom even technically successful operations may be followed by months of impaired mobility, institutional care, or enduring dependence on caregivers.
The concept of postoperative recovery trajectory is gaining importance across perioperative medicine, geriatrics, and health services research. Recovery after major surgery is not binary. Patients may recover promptly, recover slowly, or enter a prolonged state of diminished function. For clinicians counseling patients before surgery, a key unanswered question has been whether these trajectories can be predicted reliably at the point of care. Existing calculators, including those derived from the American College of Surgeons National Surgical Quality Improvement Program and similar datasets, mainly estimate short-term events such as mortality or major complications. They do not directly address long-term independence.
The present study by Jacobs and colleagues addresses this gap in a population of US veterans undergoing major surgery in Veterans Affairs hospitals. Its clinical significance is substantial. Veterans receiving surgical care through the VA are often older, have multimorbidity, and may have baseline functional limitations that make long-term recovery especially salient. A tool that predicts not only whether patients survive surgery, but whether they are likely to regain independence in a timely fashion, could materially improve informed consent, prehabilitation planning, discharge preparation, and postoperative resource allocation.
Study Design and Methods
Design and data sources
This was a retrospective cohort study using VA Surgical Quality Improvement Program data from 2016 through 2019. The investigators linked these data to multiple VA data sources to capture a broad set of candidate predictors. The use of integrated VA data is a notable strength because it allows incorporation of longitudinal clinical and utilization information beyond the immediate surgical episode.
Population and exposure
The exposure of interest was surgical care delivered in VA hospitals. The abstract identifies the cohort broadly as patients undergoing major surgery within the VA system, although the specific inclusion and exclusion criteria, procedural classes, and sample size are not reported in the abstract. The focus on veterans should be kept in mind when considering external validity, given the distinctive demographic and comorbidity profile of this population.
Outcome definition
The primary outcome was a postoperative recovery trajectory categorized into two groups: routine or slow recovery, defined as recovery within 0 to 60 days, versus protracted recovery or loss of independence, defined as lasting 6 months or more. This is a clinically meaningful reframing of postoperative outcomes. Rather than measuring isolated adverse events, the study attempts to capture the patient-centered arc of recovery over time.
Model development
The investigators evaluated a broad range of candidate predictors and used LASSO regression to develop the predictive model. LASSO, or least absolute shrinkage and selection operator, is a penalized regression approach that reduces overfitting and performs variable selection by shrinking less informative coefficients toward zero. For clinical prediction, this offers a reasonable balance between interpretability and performance.
The authors also compared the regression model with a gradient boosting approach, a tree-based machine learning method that can capture nonlinear interactions. This comparison is important because many current prediction studies invoke machine learning without establishing whether it meaningfully outperforms simpler models in real-world use.
Key Findings
Predictive performance
The study reports high discriminative performance across both training and test samples, with c-statistics ranging from 0.906 to 0.908. In clinical prediction, a c-statistic above 0.90 is typically considered excellent discrimination. This suggests the model was highly capable of distinguishing patients likely to follow a routine or slow recovery path from those at risk for a prolonged recovery course or long-term dependence.
The model was also well balanced operationally, with sensitivity of 83.0% to 83.6% and specificity of 82.4% to 82.5%. These numbers matter clinically. A highly sensitive tool reduces the chance of missing patients who may benefit from enhanced perioperative support, while reasonable specificity helps limit unnecessary allocation of scarce resources to lower-risk patients.
Most influential predictors
The strongest predictor was the Care Assessment Needs score for 90-day hospitalization. This finding is both intuitive and important. The CAN score is a VA population health tool developed to estimate risk of hospitalization or death using routinely available clinical and administrative data. Its prominence in this model suggests that broad measures of underlying vulnerability may be more informative for long-term recovery than procedure-specific variables alone.
Functional status was the next strongest predictor. This aligns with a large body of perioperative and geriatric literature showing that baseline function and frailty strongly influence outcomes after surgery. Functional dependence, difficulty with activities of daily living, and reduced physiologic reserve are consistently associated with worse recovery, institutionalization, and mortality.
Although the abstract does not list all selected covariates, the emphasis on CAN score and function suggests that the model captures a patient’s global health state, rather than focusing narrowly on intraoperative or disease-specific factors. This is appropriate for predicting a long-horizon outcome such as recovery trajectory.
Comparison with machine learning
The gradient boosting model reached a c-statistic of 0.920, only modestly higher than the LASSO model. This is a key practical finding. The marginal improvement may not justify the reduced transparency and more complex implementation of a machine learning model, especially if the goal is bedside use and shared decision-making. In many clinical settings, a slightly less performant but more interpretable model is preferable.
This result also contributes to a broader methodological point in clinical AI: sophisticated algorithms do not always yield clinically meaningful gains when high-quality structured predictors already capture much of the signal. For implementation, calibration, transparency, workflow integration, and ease of communication are often as important as incremental changes in the c-statistic.
Clinical Interpretation
The study’s major contribution is conceptual as well as technical. It shifts perioperative risk assessment away from short-term event prediction and toward outcomes that are closer to what patients actually ask in clinic: How long will it take me to recover? Will I be able to live independently? Will I need more help after surgery than I do now?
For surgeons, anesthesiologists, geriatricians, and perioperative teams, such a tool could support several decisions. First, it may improve preoperative counseling by making long-term recovery risk more concrete. Second, it may identify candidates for prehabilitation, nutritional optimization, medication review, mobility training, or social work intervention before surgery. Third, it may help target postoperative rehabilitation resources, transitional care, and caregiver support to those most likely to need them.
In the VA system specifically, where care coordination and longitudinal data infrastructure are comparatively strong, such a calculator could be embedded in the electronic record and used to trigger pathways for high-risk patients. Examples might include proactive physical therapy referral, early discharge planning, home support assessment, or geriatric comanagement.
The prominence of functional status also reinforces a clinical message that should already be familiar but is often underemphasized in surgical workflows: function is not merely a background characteristic; it is a major determinant of outcome. Measuring it systematically before surgery is therefore not optional if the goal is patient-centered perioperative care.
Strengths of the Study
The study appears to have several notable strengths. It uses a large, integrated national health system dataset, which likely improves completeness of longitudinal outcome ascertainment. It defines an outcome of direct patient relevance rather than relying solely on conventional perioperative endpoints. It also uses a transparent and pragmatic modeling strategy, then benchmarks that strategy against a more complex machine learning approach.
Another strength is the focus on implementation potential. The study does not merely demonstrate an association between frailty-related variables and recovery; it moves toward a usable calculator. Clinical prediction research is most valuable when it bridges epidemiology and bedside care, and this work makes progress in that direction.
Limitations and Cautions
Several limitations should be considered. First, the study is retrospective and based on observational data. As with all such models, performance depends on the quality and completeness of the underlying data sources, and predictor-outcome relationships may shift over time.
Second, the VA population is not fully representative of the broader US surgical population. Veterans undergoing surgery are disproportionately older and male, often with substantial medical and psychosocial comorbidity. External validation in non-VA settings, including community hospitals, academic centers, and more sex-balanced cohorts, will be essential before widespread adoption.
Third, the abstract does not provide details on calibration, net benefit, subgroup performance, or fairness across demographic strata. Excellent discrimination does not guarantee that predicted probabilities are well calibrated, and implementation requires confidence that the model performs adequately across age, race, sex, procedure type, and baseline disability burden.
Fourth, the outcome definition, while clinically attractive, may depend on how recovery and independence were operationalized in the underlying datasets. The specific measures used to determine return to baseline or persistent dependence will be important to review in the full article. Patient-reported outcomes would strengthen this work, although they are often difficult to obtain at scale.
Finally, prediction alone does not improve outcomes unless actionable interventions follow. High-risk identification must be linked to care pathways that can plausibly modify recovery trajectory.
Relation to Existing Evidence
The study is consistent with prior literature showing that frailty, disability, and diminished functional reserve are powerful predictors of adverse surgical outcomes. In older adults, postoperative functional decline may be more common and more consequential than many traditional complications. Work from geriatric surgery and palliative care has repeatedly emphasized that survival without preserved function may not align with patient goals.
Professional guidance increasingly reflects this perspective. The American College of Surgeons and the American Geriatrics Society have highlighted preoperative assessment of frailty, cognition, function, and social support in older surgical patients. Similarly, perioperative quality initiatives are moving beyond mortality to include disability-free survival, return to baseline living situation, and patient-reported recovery.
What distinguishes the present study is its attempt to operationalize these concepts into a practical prediction tool using routinely collected health system data. That translational step is crucial if long-term functional outcomes are to become part of standard perioperative risk communication rather than remaining a research abstraction.
Practice Implications
If externally validated, this calculator could change perioperative conversations in several ways. For a patient considering major surgery, a discussion framed only around operative mortality and major complications may underestimate the lived impact of the procedure. Adding an individualized estimate of prolonged recovery or loss of independence may lead some patients to pursue surgery with better preparation, and others to reconsider the balance of benefits and burdens.
From a systems perspective, the calculator could support resource stewardship. High-risk patients might benefit from targeted interventions such as prehabilitation, delirium prevention bundles, geriatric consultation, rehabilitation planning, caregiver education, or closer post-discharge follow-up. Conversely, lower-risk patients could avoid unnecessary high-intensity pathways.
Importantly, risk communication must be careful and nuanced. A prediction of increased risk should not be framed as determinism. Rather, it should support shared decision-making, clarify expectations, and prompt efforts to mitigate modifiable contributors to poor recovery.
Future Directions
Several next steps follow naturally from this work. External validation is the immediate priority, especially across procedure types and non-VA populations. Calibration analyses and decision-curve analyses would help clarify clinical usefulness. Prospective implementation studies are also needed to determine whether use of the calculator changes conversations, care plans, or outcomes.
Another important direction is integration with patient-reported outcomes. Recovery and independence are deeply personal experiences, and future models may be strengthened by combining administrative predictors with direct assessments of mobility, symptom burden, cognition, and social support.
Finally, interventional studies should test whether identifying high-risk patients leads to better outcomes when paired with tailored perioperative programs. Prediction is valuable, but prevention and recovery optimization are the ultimate goals.
Conclusion
Jacobs and colleagues report a clinically important advance in perioperative prediction science: a high-performing VA-based risk calculator for protracted postoperative recovery and long-term loss of independence after major surgery. The model performed strongly, with c-statistics above 0.90 and balanced sensitivity and specificity, and it identified the Care Assessment Needs score and functional status as dominant predictors.
The findings reinforce a central truth of modern surgical care: long-term function matters, and baseline vulnerability heavily shapes recovery. Although external validation and implementation research are still needed, this study offers a credible path toward risk assessment that is more aligned with what patients care about than traditional 30-day endpoints alone.
Funding and Registration
The abstract does not report funding information or a ClinicalTrials.gov registration number. As a retrospective cohort study using existing VA data, trial registration may not have been applicable. Readers should consult the full publication for funding disclosures and conflicts of interest.
Citation
Jacobs MA, Maxwell JH, Cashy J, Boudreaux-Kelly MY, McCoy JL, Intrator O, Kinosian B, Youk A, Shireman PK, Hall DE. Predicting Protracted Recovery and Long-Term Loss of Independence after Major Surgery in Veterans. Annals of Surgery. 2026-05-05. PMID: 42083088. Available at: https://pubmed.ncbi.nlm.nih.gov/42083088/
Selected Contextual References
Hall DE, Arya S, Schmid KK, et al. Development and initial validation of the Risk Analysis Index for measuring frailty in surgical populations. JAMA Surgery. 2017;152(2):175-182.
American College of Surgeons National Surgical Quality Improvement Program and American Geriatrics Society. Optimal preoperative assessment of the geriatric surgical patient: a best practices guideline. Accessed via ACS/AGS guidance documents.
McIsaac DI, Moloo H, Bryson GL, van Walraven C. The association of frailty with outcomes and resource use after emergency general surgery: a population-based cohort study. Anesth Analg. 2017;124(5):1653-1661.
Berian JR, Zhou L, Hornor MA, et al. Optimizing surgical quality datasets to care for older adults: lessons from the American College of Surgeons NSQIP Geriatric Surgery Pilot. J Am Coll Surg. 2017;225(6):702-712.
