Risk adjustment and observation time: comparison between cross-sectional and 2-year panel data from the Medical Expenditure Panel Survey (MEPS).
Bottom Line:
The results of this technique were not satisfying.One reason could be that the length of time to document consumption might be associated with the mean and variance of observed health care consumption.The reason is not clear and we will continue studying this phenomenon.
View Article:
PubMed Central - PubMed
Affiliation: School of Public Health, University of Alberta, Edmonton, Alberta T6G2T4 Canada.
ABSTRACT
Background: Risk adjustment models were used to estimate health care consumption after adjusting for individual characteristics or other factors. The results of this technique were not satisfying. One reason could be that the length of time to document consumption might be associated with the mean and variance of observed health care consumption. This study aims to use a simplified mathematical model and real-world data to explore the relationship of observation time (one or two years) and predictability. Methods: This study used cross-sectional (one-year) and 2-year panel data sets of the Medical Expenditure Panel Survey (MEPS) from 1996 to 2008. Comparisons of the health care consumption (total health expenditure, emergency room (ER) and office-based visits) included ratios of means and standard errors (SEs). Risk adjustment models for one- and two-year data used generalized linear model. Results: The ratios of mean health care consumption (two-year to one-year total expenditure, ER and office-based visits) seemed to be two in most age groups and the ratios of SEs varied around or above two. The R-squared of two-year models seemed to be slightly better than that of one-year models. Conclusions: We find health expenditure and ER or office-based visits observed in two consecutive years were about twice those observed in a single year for most age, similar to the ratios predicted in mathematical examples. The ratios of mean spending and visits varied across age groups. The other finding is that the predictability of two-year consumption seems better than that of one-year slightly. The reason is not clear and we will continue studying this phenomenon. No MeSH data available. Related in: MedlinePlus |
Related In:
Results -
Collection
License 1 - License 2 getmorefigures.php?uid=PMC4340859&req=5
Mentions: In Figure 2, the ratios of mean spending and standard errors (SEs) were plotted against ages. The ratios of mean spending seem to align with the horizontal line of two, except for lower at age zero, suggesting the spending sum at age zero and one was less than twice the amount at age zero. Most of the SE ratios were above two, except for certain ages. The mean ratios of mean spending and SEs were 2.09 and 2.58.Figure 3 presents the mean emergency room (ER) visits by age. The distribution is not similar to that of health spending. The ER visits were more at age 1 and around age 20. After age 30, ER visits decreased until age 60 to 70. In Figure 4, the ratios of mean visits and SEs were plotted against ages. The mean ratios of mean ER visits and SEs 2.02 and 2.18.Figure 5 presents the mean office-based visits by age and the J-shape distribution is similar to that of health spending. In Figure 6, the ratios of means and SEs were mostly greater than two. The mean ratios of mean office-based visits and SEs were 1.99 and 2.42.Figure 7 presents the estimated R-squared predicted by basic demographic information (gender, race and regions) of all ages and those predicted by basic and other variables from age 45 to 65 years. Because the R-squared from 2-year spending of all ages seemed to be higher of most age groups, the health spending observed in two years might be more predictable than that observed in one year. Figures 8 and9 present the R-squared predicted from ER and office-based visits. In general, the R-squared was larger if predicted with more variables. The predictability (R-squared) of health spending and office-based visits seemed to be higher at age 20 to 30 years.Figure 1 |
View Article: PubMed Central - PubMed
Affiliation: School of Public Health, University of Alberta, Edmonton, Alberta T6G2T4 Canada.
Background: Risk adjustment models were used to estimate health care consumption after adjusting for individual characteristics or other factors. The results of this technique were not satisfying. One reason could be that the length of time to document consumption might be associated with the mean and variance of observed health care consumption. This study aims to use a simplified mathematical model and real-world data to explore the relationship of observation time (one or two years) and predictability.
Methods: This study used cross-sectional (one-year) and 2-year panel data sets of the Medical Expenditure Panel Survey (MEPS) from 1996 to 2008. Comparisons of the health care consumption (total health expenditure, emergency room (ER) and office-based visits) included ratios of means and standard errors (SEs). Risk adjustment models for one- and two-year data used generalized linear model.
Results: The ratios of mean health care consumption (two-year to one-year total expenditure, ER and office-based visits) seemed to be two in most age groups and the ratios of SEs varied around or above two. The R-squared of two-year models seemed to be slightly better than that of one-year models.
Conclusions: We find health expenditure and ER or office-based visits observed in two consecutive years were about twice those observed in a single year for most age, similar to the ratios predicted in mathematical examples. The ratios of mean spending and visits varied across age groups. The other finding is that the predictability of two-year consumption seems better than that of one-year slightly. The reason is not clear and we will continue studying this phenomenon.
No MeSH data available.