Summary
Starfield and Kinder argue that condition count — the dominant method in epidemiological multimorbidity research — is a poor predictor of health resource use. Using validated applications of the Johns Hopkins Adjusted Clinical Groups (ACG) system, they show that what matters is the type of illness burden (captured across 32 Aggregated Diagnosis Groups), not the raw number of chronic conditions. Morbidity burden assessed this way outperforms condition counts in explaining resource use, predicts stable multi-year trajectories, and reveals non-random morbidity clustering that is systematically more severe in disadvantaged populations.
Key Claims
- Type over count: The number of different types of illness — coded via 32 Aggregated Diagnosis Groups (ADGs) → 12 Collapsed ADG (CADG) categories → ~100 mutually exclusive ACG combinations — is a substantially better predictor of resource use than a raw count of chronic conditions.
- Resource use paradox: The majority of individuals with any single common chronic condition have below-average resource use: ~80% of those with hypertension, ~70% with hyperlipidemia, ~40% with diabetes, and ~33% with osteoporosis use less than average resources. High users are disproportionately those with multiple types of illness across ADG categories.
- Morbidity burden > sum of disease costs: Using British Columbia, Canada administrative data, resource use increases more than linearly with morbidity burden at every stratum of chronic condition count (0–4+), demonstrating that total morbidity burden is not reducible to the sum of individual disease costs.
- Non-random morbidity clustering: Using Manitoba data, ACG morbidity burden scores correlate with socioeconomic status (SES) in expected directions — disadvantaged populations carry systematically higher morbidity burdens, confirming that morbidity is not randomly distributed.
- Trajectory stability: In Taiwan's National Health Insurance program, ~80% of the population remains in the same or an adjacent ACG morbidity burden trajectory group over multiple years — morbidity trajectories are stable and predictive, not random year-to-year variation.
- ACG System applications: Beyond research, the ACG System is used operationally for population profiling, provider profiling, equitable budget allocation (Sweden, Spain), high-risk patient identification (Germany, US, UK), and monitoring policy reforms; as of 2011 it operates in the US, Canada, Sweden, Spain, UK, Germany, Israel, South Africa, and Malaysia.
Concepts Introduced or Extended
- Multimorbidity — extends with ACG type-based measurement framework; resource use paradox; morbidity trajectory stability
Entities Mentioned
Quotes
"The number of different types of illness is a better indicator of overall morbidity burden than the number of different chronic conditions."
"These analyses using the ACG System make it clear that morbidity is not randomly distributed in the population."
"Resource use increased more than linearly with increasing morbidity burden, at every stratum of number of chronic conditions (0–4 or more)."
My Take
This is methodological advocacy, not an original empirical study. Starfield and Kinder make no new estimates — every result cited is drawn from prior ACG validation studies (Israel, Manitoba, British Columbia, Taiwan). The paper's value is in synthesizing the type-based measurement case, providing vocabulary (ADG, CADG, ACG) that appears throughout applied multimorbidity literature, and demonstrating real-world policy uptake. The resource-use paradox finding — that most people with any single chronic condition are below-average resource users — is a powerful corrective to condition-count assumptions and the most useful substantive contribution to Multimorbidity. The paper has a notable disclosure: Johns Hopkins University (JHU) holds copyright on ACG software and collects royalties from insurance plans; the authors are JHU faculty. This does not invalidate the findings but should be noted for the advocacy framing.