By: Chad E Cook, PT, PhD, FAPTA
Background: In 2011, I was part of a team of researchers who looked at very large datasets and routinely used publicly available risk adjustment measures. We had access to the Nationwide Inpatient Sample, which is the largest publicly available all‑payer inpatient healthcare database in the United States and remains widely used for research, policy analysis, and benchmarking. It provides a powerful means of studying hospital utilization, costs, outcomes, and trends at the national level. We compared a novel risk adjustment measure (the condition-specific measure) with a well-known risk adjustment tool (the Deyo index) and found that our novel tool was superior, albeit still flawed. In the discussion, we state (paraphrased) “although risk adjustment measures have been used in outcomes research for decades, and attempt to correct for case severity through statistical methods, there are no risk-adjustment approaches that can control for every factor affecting outcomes of care” [1].
We were right then, and the point still holds today: no single risk‑adjustment method can fairly compare fundamentally different institutions, such as a Level 1 trauma center and a community hospital. It’s an apples‑to‑oranges comparison problem.
But what if someone could compare apples to apples, and separately, oranges to oranges, and control for patient differences across those distinct groups? Good news, one can. The purpose of this blog is to define both risk stratification and risk adjustment and discuss when to use each method.
Risk stratification: Risk stratification is the process of pre-emptively sorting individuals into meaningful groups based on their likelihood of experiencing a particular outcome. Think of it as creating tiers, such as low, medium, and high risk, so one can target interventions or resources more effectively. Risk stratification has been used prospectively to guide care based on stratification sequences [2] and retrospectively to group individuals when evaluating overall outcomes [3,4].
When evaluating outcomes retrospectively, risk stratification is essential for accurately interpreting results in populations in which known confounding factors constrain improvement or recovery. Certain patient characteristics exert such a strong influence on outcomes that failing to stratify can obscure true treatment effects or lead to misleading comparisons. Social risk is a particularly powerful example [5,6]. Factors such as socioeconomic disadvantage, limited social support, housing instability, or low health literacy have been shown to profoundly shape recovery trajectories, such that traditional adjustment alone may not fully account for their impact. In these situations, examining outcome scores within defined strata of social risk, for example, comparing patients with two or more social risk factors separately from those with three or more, provides a clearer, more equitable understanding of performance and outcome variation. Stratification in this way ensures that analyses reflect meaningful comparisons among patients with similar levels of social complexity, rather than masking disparities by averaging across fundamentally different groups.
Risk adjustment: Risk adjustment, in contrast, functions as a statistical correction mechanism. Its purpose is to account for differences in underlying patient risk profiles (e.g., age, comorbidities, socioeconomic status, disease severity, or prior utilization) so that comparisons between groups, providers, health plans, or care models are genuinely fair [7,8]. Without this correction, any observed differences in outcomes could simply reflect that one non-randomized group treated sicker, older, or more medically complex patients rather than delivering better or worse care.
Risk adjustment models incorporate important baseline characteristics into the analysis, allowing researchers to isolate the effect of the factor they truly want to study. By controlling for case‑mix variation, risk adjustment helps ensure that performance metrics, cost comparisons, and outcome evaluations reflect differences in care, not differences in the populations being served. This is especially important in large datasets, where even small imbalances in patient characteristics can distort results at scale. In essence, risk adjustment statistically neutralizes the influence of factors outside a provider’s control. This makes it beneficial for evaluating quality, benchmarking performance, setting reimbursement rates, and conducting rigorous health services research. By itself it can be very helpful……to a point…..[1,7,9]
When to use risk stratification, risk adjustment or both: Risk stratification and risk adjustment serve different, but complementary purposes when working with large datasets (Table 1) [6-8]. Risk stratification is most useful when the goal is to categorize individuals into meaningful risk groups to guide decision‑making, target interventions, or understand how risk is distributed across a population. It answers the question: Who is at higher or lower risk? In contrast, risk adjustment is essential when comparing outcomes across groups, providers, or systems where underlying patient differences could bias results. It statistically controls for variations in case mix so comparisons reflect differences in care rather than differences in patient complexity.
A Simple Way to Remember the Difference
- Risk stratification involves determining who is at risk of a poorer outcome.
- Risk adjustment involves the question of how to adjust and compare outcomes fairly.
Many analyses benefit from using both. One can use stratification to segment the population into comparable subgroups, and adjustment to ensure fair comparisons within or across those strata. Together, they allow researchers to compare “apples to apples” while still understanding how risk varies across the dataset.
Table 1. When to use risk stratification or risk adjustment methods.
| Feature | Risk Stratification | Risk Adjustment |
| Primary Purpose | Categorizes individuals into risk groups | Statistically controls for baseline differences |
| Key Question Answered | Who is at risk? | Are comparisons fair? |
| Typical Use Case | Targeting interventions, resource allocation, population segmentation | Comparing outcomes across providers, plans, or groups |
| Analytic Role | Classification | Correction |
| Level of Application | Individual or subgroup level | Group or system level |
| Common Inputs | Clinical factors, psychosocial factors, utilization history | Demographics, comorbidities, severity, socioeconomic variables |
| Output | Risk tiers (e.g., low/medium/high) | Adjusted rates, risk‑standardized outcomes |
| Strengths | Supports operational decisions and care management | Ensures fair benchmarking and performance evaluation |
| When Used Together | Stratify population first, then adjust outcomes within or across strata | Adjusts for the most accurate and equitable comparisons |
Summary: Working with large datasets in healthcare, insurance, or population analytics often entails navigating two closely related yet fundamentally distinct concepts: risk stratification and risk adjustment. Both help you make sense of complex populations, yet they serve distinct purposes and answer different questions. Knowing when to use each one can dramatically improve the clarity and impact of your analysis.
Layperson’s Summary: Everyone is different, and it is unfair and inaccurate to compare individuals with injuries or pathologies as if they should have similar outcomes. Scientists recognize this and have developed risk stratification and risk adjustment methods to improve the accuracy of comparisons among patients who receive similar treatments. This blog discusses both approaches and advocates their use when working with large datasets.
Portions of this blog were developed with assistance from AI‑based writing tools (primarily grammar and structure) and were reviewed and edited by the author for accuracy and clarity.
References
- Goode AP, Cook C, Gill JB, Tackett S, Brown C, Richardson W. The risk of risk-adjustment measures for perioperative spine infection after spinal surgery. Spine (Phila, Pa. 1976). 2011 Apr 20;36(9):752-8.
- Hill JC, Whitehurst DG, Lewis M, Bryan S, Dunn KM, Foster NE, Konstantinou K, Main CJ, Mason E, Somerville S, Sowden G, Vohora K, Hay EM. Comparison of stratified primary care management for low back pain with current best practice (STarT Back): a randomised controlled trial. Lancet. 2011 Oct 29;378(9802):1560-71.
- Buttar A, Fazal-Ur-Rehman M, Manes T, Turnow M, Williamson TK, Taylor BC, Weick JW, Bowers C. Orthopedic frailty risk stratification (OFRS): a systematic review of the frailty indices predicting adverse outcomes in orthopedics. J Orthop Surg Res. 2025 Mar 6;20(1):247.
- Sequeira SB, Scuderi GR, Mont MA. Patient Frailty is an Important Metric to Predict Outcome After Revision Arthroplasty Procedures. J Arthroplasty. 2024 May;39(5):1149-1150.
- Bernstetter A, Brown NH, Fredhoff B, Rhon DI, Cook C. Reporting and incorporation of social risks in low back pain and exercise studies: A scoping review. Musculoskelet Sci Pract. 2025 Jun;77:103310.
- National Academies of Sciences, Engineering, and Medicine. Accounting for Social Risk Factors in Medicare Payment: Identifying Social Risk Factors. Washington, DC: National Academies Press; 2016.
- Iezzoni LI. Risk Adjustment for Measuring Health Care Outcomes. 4th ed. Chicago, IL: Health Administration Press; 2013.
- Ash AS, Ellis RP. Risk adjustment and the future of health care. Health Care Financ Rev. 2012;33(4):1‑9.
- Zaslavsky AM. Statistical issues in reporting quality data: small samples and case mix variation. Int J Qual Health Care. 2001;13(6):481 488.
