Summary
Using a high-dimensional Bayesian multivariate probit model (5,356 unique correlation parameters) fitted to a 10% random sample of 2009 disability insurance (DI) initial claimants (n=157,835), Meseguer estimates pairwise correlations between the 100 most common primary diagnoses among qualified applicants. The paper demonstrates that secondary diagnosis codes carry substantial predictive value beyond primary codes, that comorbidity rates have risen sharply since 1997, and that commonly assumed comorbidity patterns — especially between mental and musculoskeletal disorders — are empirically wrong. The mental-musculoskeletal correlation is negative, not positive, at the initial-claim level.
Key Claims
- Comorbidity rising: 56.3% of 1997 claimants had a secondary diagnosis; 71.1% of 2009 claimants (71.4% by 2010). The secondary code is increasingly the norm, not the exception.
- Mental disorders systematically undercounted by primary code: Affective/mood disorders appear as secondary for 22.9% of claimants but as primary for only 14.2%. Anxiety disorders appear as secondary (8.76%) far more often than as primary (2.37%). Disorders of the back, by contrast, appear mostly as primary. Any analysis using only primary codes understates the mental disorder burden in the applicant pool.
- Mental-musculoskeletal myth refuted: The two most common diagnoses — affective/mood disorders and disorders of the back — are negatively correlated. They combine less often than their overall frequencies predict. Affective/mood disorders also show negative correlations with osteoarthrosis, muscle/ligament disorders, diffuse connective tissue diseases, other bone/cartilage diseases, and curvature of the spine. Disorders of the back show negative correlations with schizophrenia, anxiety, personality disorders, organic mental impairments, and substance addiction.
- Neurological cluster: Five diagnostic groups exhibit strong positive inter-group correlations with one another: (1) organic mental disorders; (2) intracranial (traumatic brain) injury; (3) malignant and benign brain neoplasms; (4) late effects of cerebrovascular disease; (5) nervous system/sense organ disorders (epilepsy, migraine, cerebral degenerations, visual disturbances). These form a coherent brain-injury/damage cluster that spans traditional diagnostic group boundaries.
- Two distinct mental subgroups (negatively correlated with each other): Group 1 — affective/mood disorders, anxiety, personality disorders, schizophrenia, substance addiction (mood-anxiety-psychotic cluster). Group 2 — autism, intellectual disability, borderline intellectual functioning, learning/attention deficit disorders (early developmental cognitive cluster). The two groups are often negatively correlated with each other despite both being classified as "mental disorders."
- Concurrent DI/SSI filing: 52.5% of qualified 2009 DI claimants applied concurrently to Supplemental Security Income (SSI). Concurrent applicants are ~8 years younger (median), have an initial allowance rate of 26.9% vs. higher for DI-only applicants, and dominate mental disorder diagnoses while underrepresenting cancers and nervous system cases.
- Secondary diagnosis predictive value: Using a naïve Bayes classifier: primary diagnosis misclassification rate 29.6%; secondary diagnosis alone 34.0% — better than state of origin, sex, ethnicity, earnings, concurrent status, reapplicant status, or education. Combined primary + secondary: 27.1%; all variables combined: 23.9%.
- Full model accuracy: 72.6% of initial determinations correctly predicted (53.6% of allowances, 84.3% of denials). Zeroing out estimated correlations degrades accuracy to 66.9%, confirming comorbidity structure adds real predictive content.
Concepts Introduced or Extended
Entities Mentioned
Quotes
"Ignoring the secondary diagnoses clearly leaves exploitable information on the table."
"The correlation estimates contradict presumptive expectations of high comorbidity between mental and musculoskeletal diagnoses. Affective/mood disorders have negative estimated correlations with many of the musculoskeletal diagnoses."
My Take
The most practically important finding is the refutation of the mental-musculoskeletal comorbidity presumption. This matters because policy discussions about DI applicants often assume that the rising share of mental disorder diagnoses is partly an artifact of applicants with musculoskeletal conditions also receiving a mental diagnosis — a "stacking" hypothesis. The negative correlation evidence undercuts this: back disorder claimants are not particularly likely to also have a mental disorder diagnosis, and vice versa. The two diagnostic groups represent distinct applicant populations. The two-subgroup finding within mental disorders (mood-anxiety-psychotic vs. developmental cognitive) is also policy-relevant because the two clusters respond differently to vocational rehabilitation and employment interventions. The concurrent DI/SSI filing statistic (52.5%) is striking — it is rarely cited in the DI literature — and implies that much of what appears as "DI-only" analysis actually excludes roughly half of the applicant pool. The growing comorbidity trend (56% → 71% with secondary codes from 1997 to 2010) also suggests that disability burden is increasing in a multi-morbidity sense beyond what primary codes capture.