Medical Costs Attributable to Diagnosed Vision Disorder and Undiagnosed Visual Loss

We calculated overall and per payer medical costs using a two-part generalized linear on 2003–2008 MEPS pooled event file data.[18, 19] MEPS is a nationally representative panel survey conducted on a sample of National Health Interview Survey respondents to capture additional detail including medical costs. In MEPS, respondents are queried on their medical history and medical care received in the past year. In the Medical Provider component, MEPS contacts the respondents’ medical providers to confirm visits, events and costs. MEPS differentiates costs by payer. Based on medical history and treatment, MEPS assigns 3-digit ICD-9 diagnosis codes to individual respondents in the Medical Conditions File. MEPS does not include optometrists in the Medical Provider component. Respondents are asked to report visits to optometrists and their respective costs, and report the costs for vision aids (glasses and contact lenses). MEPS does not confirm these costs. Due to this structure, the cost of care provided by optometrists may not be included in the medical provider component and may not be associated with diagnosed conditions. Thus, we calculate the medical costs attributable to diagnosed disorders and undiagnosed disorders on total medical costs excluding costs for optometry services and vision aids, which we calculate separately.

Table 4.1. Medical Costs Attributable to Diagnosed Disorders

Medical Costs Attributable to Diagnosed Disorders

Table 4.2. Medical Costs Attributable to Undiagnosed Visual Loss

Medical Costs Attributable to Undiagnosed Visual Loss
*Not statistically different from zero

The first part of the two-part model used a logistic regression to predict the probability of positive expenditures. The dependent variable in the second part is defined as annual total medical expenditures excluding optometry expenses, which include the cost of medical vision aids (e.g., glasses, contact lenses) and optometrist visits. Using a generalized linear model with a gamma distribution and a log link, the second part of the model predicts expenditures conditional on having positive expenditures. Multiplying the predicted probability from the first part of the model with predicted (conditional) expenditures from the second part, we generated predicted annual expenditures for each individual in the data as a function of the individual’s condition profile and demographics.[19]

Both parts of the model included the same set of independent variables. These included dichotomous variables identifying the comprehensive vision disorder variable, a variable representing self-reported difficulty seeing in the absence of a diagnosed disorder (undiagnosed visual loss), and expensive comorbidities: hypertension and diabetes. All regressions also included as independent variables the following demographic characteristics: age, age squared, sex, race/ethnicity (white non-Hispanic, other), education (missing degree, younger than 17 years of age, no degree, high school diploma, college degree, graduate degree), region (northeast, midwest, south, west), insurance status (private, public, uninsured), marital status, family size, family income (<100%, 100% to 199%, 200% to 399%, >400% of the poverty line), and year.[20] Bootstrapped standard errors that account for the sampling design of MEPS were generated for all estimates. Medical costs of diagnosed disorders and undiagnosed vision loss, excluding the cost of optometry visits and medical vision aids, are shown in Tables 4.1 and 4.2, respectively.

Non-optometry medical costs per person attributable to each disease were calculated using the following method that minimizes double-counting of expenditures across diseases:[19]

  1. Every unique combination of the chronic diseases listed among independent variables observed in the data was identified.
  2. Expenditures were predicted for each individual.
  3. For each unique combination of diseases, we subtracted from Step 2 the predicted expenditures for an otherwise identical person without the combination of diseases. This provides an estimate of the costs attributable to every unique combination of diseases.
  4. The coefficients of the diseases from the second part of the two-part econometric model were used as importance weights to redistribute costs associated with jointly occurring diseases to constituent diseases (e.g., to redistribute the costs of vision disorders with hypertension back to vision disorders and hypertension separately).
  5. The application averages the redistributed costs over the population with each disease.