Scholarship and Significance
-
Posted by Robert Kaestner
- Posted on 6/26/2024 12:23:43 PM
The article by Drs. Alexander and Schnell purports to show that: "We further find that higher physician reimbursement leads to more office visits, better self-reported health, and reduced school absenteeism among the program’s beneficiaries."
This conclusion seems like an overstatement.
Here are the results with a few comments:
1. Table 3:
a. Among adults on Medicaid, a $10 increase in Medicaid reimbursement for new patient is associated with a statistically significant (p=0.045), but small, 2% increase in probability of good or excellent health among adults.
In event study Figure 4, there appears to be a violation of parallel trends although not significant for this outcome because of a lack of statistical power. Two of the three estimates in pre-period are as large as the two post-period estimates. So, even this tiny effect should be interpreted cautiously.
b. Among adults on Medicaid, a $10 increase in Medicaid reimbursement for new patient is associated with an insignificant 1.5% decrease in the probability of being in poor health.
c. Among adults on Medicaid, a $10 increase in Medicaid reimbursement for new patient is associated with an insignificant 1% increase in recent office visit.
d. Among adults with private insurance, a $10 increase in Medicaid has no statistically significant effects on outcomes.
In sum, at most, a small effect on probability of being in excellent health with some caution regarding validity of research design.
a. Among children on Medicaid, a $10 increase in Medicaid reimbursement for new patient is associated with a statistically significant 0.5 percentage point (25%) decrease in probability of having trouble finding a MD.
The baseline mean is 2.2%--basically almost no children have a problem and raising fees will have little effect on a non-problem. Why even study this non-problem?
b. Among children with private insurance, a $10 increase in Medicaid reimbursement for new patient is associated with a statistically significant 0.2 percentage point (19%) increase in probability of having trouble finding a MD.
The baseline mean is 0.8%--virtually no children have a problem. Why study it?
c. Among children on Medicaid or private insurance, a $10 increase in Medicaid reimbursement for new patient is associated with no statistically significant changes in school attendance.
In Table 5, using NAEP, a statistically significant 3% decrease in probability of missing 3+ days of school is reported for those in grade 4. No effects on those in grade 8. Figure A.7 suggests a violation of parallel trends for those in grade 4.
An omission in child analysis seems to be not examining child health? Why not study health as was done for adults? There are also many other indicators of child health that could be used, for example, asthma episode and/or admission; hospitalization; and emergency department use.
Medicare Payment Rates:
Dr.'s Alexander and Schnell suggest that Figure A1, panel B shows that changes in payments for new patients is highly correlated with changes in payments for established patients. It is hard to assess this claim, but there does seem to be substantial divergence. Established patient visits dominate physicians' effort and account for a large share of revenue. Dr.'s Alexander and Schnell mention that results are robust to use of different payment rates, but do not show results. Given the marginally significant, small and inconsistent results, it would be helpful to reader to show these results.
It is also known that even within a state and CPT code there is substantial variation in Medicaid reimbursement. From Chen J, van den Berghe E, Kaestner R. Is Primary Care A Substitute or Complement for Other Medical Care? Evidence from Medicaid. Forum Health Econ Policy. 2019:
"One interesting, and, as far as we are aware, previously undocumented aspect of Medicaid reimbursement is that there is substantial variation of fees within a CPT code, state and year. Appendix Table 12 illustrates this point. It shows the distribution of fees for one CPT code (99213) by state-year. The table reports the mean fee, modal fee, share (percent) of observations with the modal fee, as well as some other measures of dispersion. Surprisingly, FFS reimbursement for the same CPT code in a state-year often has substantial variation. For example, as shown, the share of observations with the modal fee ranges from 0.33 to 1.0, and standard deviations are sometimes in the teens. This variation in Medicaid fees is largely ignored in other studies because of how the fee information is collected."
Scholarship and Significance
This conclusion seems like an overstatement.
Here are the results with a few comments:
1. Table 3:
a. Among adults on Medicaid, a $10 increase in Medicaid reimbursement for new patient is associated with a statistically significant (p=0.045), but small, 2% increase in probability of good or excellent health among adults.
In event study Figure 4, there appears to be a violation of parallel trends although not significant for this outcome because of a lack of statistical power. Two of the three estimates in pre-period are as large as the two post-period estimates. So, even this tiny effect should be interpreted cautiously.
b. Among adults on Medicaid, a $10 increase in Medicaid reimbursement for new patient is associated with an insignificant 1.5% decrease in the probability of being in poor health.
c. Among adults on Medicaid, a $10 increase in Medicaid reimbursement for new patient is associated with an insignificant 1% increase in recent office visit.
d. Among adults with private insurance, a $10 increase in Medicaid has no statistically significant effects on outcomes.
In sum, at most, a small effect on probability of being in excellent health with some caution regarding validity of research design.
a. Among children on Medicaid, a $10 increase in Medicaid reimbursement for new patient is associated with a statistically significant 0.5 percentage point (25%) decrease in probability of having trouble finding a MD.
The baseline mean is 2.2%--basically almost no children have a problem and raising fees will have little effect on a non-problem. Why even study this non-problem?
b. Among children with private insurance, a $10 increase in Medicaid reimbursement for new patient is associated with a statistically significant 0.2 percentage point (19%) increase in probability of having trouble finding a MD.
The baseline mean is 0.8%--virtually no children have a problem. Why study it?
c. Among children on Medicaid or private insurance, a $10 increase in Medicaid reimbursement for new patient is associated with no statistically significant changes in school attendance.
In Table 5, using NAEP, a statistically significant 3% decrease in probability of missing 3+ days of school is reported for those in grade 4. No effects on those in grade 8. Figure A.7 suggests a violation of parallel trends for those in grade 4.
An omission in child analysis seems to be not examining child health? Why not study health as was done for adults? There are also many other indicators of child health that could be used, for example, asthma episode and/or admission; hospitalization; and emergency department use.
Medicare Payment Rates:
Dr.'s Alexander and Schnell suggest that Figure A1, panel B shows that changes in payments for new patients is highly correlated with changes in payments for established patients. It is hard to assess this claim, but there does seem to be substantial divergence. Established patient visits dominate physicians' effort and account for a large share of revenue. Dr.'s Alexander and Schnell mention that results are robust to use of different payment rates, but do not show results. Given the marginally significant, small and inconsistent results, it would be helpful to reader to show these results.
It is also known that even within a state and CPT code there is substantial variation in Medicaid reimbursement. From Chen J, van den Berghe E, Kaestner R. Is Primary Care A Substitute or Complement for Other Medical Care? Evidence from Medicaid. Forum Health Econ Policy. 2019:
"One interesting, and, as far as we are aware, previously undocumented aspect of Medicaid reimbursement is that there is substantial variation of fees within a CPT code, state and year. Appendix Table 12 illustrates this point. It shows the distribution of fees for one CPT code (99213) by state-year. The table reports the mean fee, modal fee, share (percent) of observations with the modal fee, as well as some other measures of dispersion. Surprisingly, FFS reimbursement for the same CPT code in a state-year often has substantial variation. For example, as shown, the share of observations with the modal fee ranges from 0.33 to 1.0, and standard deviations are sometimes in the teens. This variation in Medicaid fees is largely ignored in other studies because of how the fee information is collected."