Medical Progress Report
No. 2 October 2005
Older Drugs, Shorter Lives? An Examination of the Health Effects of the Veterans Health Administration Formulary
Frank R. Lichtenberg
Columbia University and National Bureau of Economic Research
1. An Institute of Medicine committee agreed to assist Congress with this review, in part because the committee saw in the VHA example an opportunity to understand and anticipate problems that all publicly funded programs are likely to encounter in this new age of pharmaceuticals. Congress asked the committee to review the restrictiveness of the National Formulary, its impact on the costs and quality of care in the VHA, and how it compared with formularies and drug-management practices in the private sector and in other public programs, especially Medicaid. Further, it found that the "VA National Formulary was not overly restrictive, and the limited available evidence suggests that it has probably meaningfully reduced drug expenditures without demonstrable adverse effects on quality." However, the committee also concluded that there were "manifold opportunities to improve the management of the formulary system used by the VHA," i.e., that the National Formulary lacked systems to ensure that: (1) new drugs are expeditiously reviewed for inclusion; (2) access to medically necessary exceptions to the formulary is consistently in place systemwide; (3) therapeutic interchange is accomplished in a flexible and consistent way, sensitive to patient risks, across the far-flung VHA system; and (4) views of critical constituencies of providers and patients are represented in the management of the National Formulary.
2. The list of drugs on the National Formulary is readily available (http://www.vapbm.org/PBM/natform.htm). However, lists of drugs on only a few of the VISN formularies are available (see, e.g. http://www.visn20.med.va.gov/webRx/rxbyname.html), and these are not in a uniform format.
3. IOM Report, 50. Blumenthal, David, and Roger Herdman, eds. (2000), Description and Analysis of the VA National Formulary, VA Pharmacy Formulary Analysis Committee, Division of Health Care Services (Washington: National Academy Press) <http://www.nap.edu/catalog/9879.html>
4. VHA Directive, 97-047. Veterans Health Administration, Department of Veterans Affairs, Washington, DC 20420, July 24, 2001, VHA DIRECTIVE 2001-044, http://www.vapbm.org/directive/vhadirective.pdf.
5. Although the final authority was vested initially in a VA PBM Executive Steering Board made up of officials from various units of the VHA central office, this board never became operational.
6. Lyles et al., 1997; Massachusetts Outpatient Formulary Guide. 1999; see also VA drug-class reviews at http://www.dppm.med.va.gov/newsite/reviews.html.
7. See http://www.fda.gov/cder/reports/rptntn98.pdf.
8. See http://www.fda.gov/cder/drugsatfda/datafiles/default.htm.
9. The National Drug File (http://www.vapbm.org/natform/NDF0305.EXE) contains data on specific products (identified by National Drug Code [NDC]). Each record includes a National Formulary indicator (YES or NO) and the name of the generic drug to which the NDC corresponds. I considered a generic drug to be on the formulary if any product corresponding to that drug was on the formulary. The fraction of products listed on the formulary is smaller than the fraction of drugs listed on the formulary. For example, only a subset of a drug's dosage forms and strengths may be listed on the formulary.
10. See http://www.fda.gov/cder/rdmt/default.htm.
11. Drawn from a nationally representative subsample of households that participated in the prior year's NCHS National Health Interview Survey. The objective is to produce annual estimates for a variety of measures of health status, health-insurance coverage, health-care use and expenditures, and sources of payment for health services. Statisticians and researchers use these data to generalize to people in the civilian noninstitutionalized population of the United States.
12. The other payers are: self or family; Medicare; Medicaid; private insurance; Champus/Champva; other federal, state and local government; workers' comp; other insurance; other private payers; and other public payers.
13. I defined the following three variables:
AGE_LT_5i = 1 if the age of prescription i was less than 5 years
= 0 otherwise
AGE_LT_10i = 1 if the age of prescription i was less than 10 years
= 0 otherwise
AGE_LT_15i = 1 if the age of prescription i was less than 15 years
= 0 otherwise
14. Although the VA National Formulary was launched in 1997, it may not have been fully implemented right away. To allow for this possibility, I compare VA with non-VA prescriptions beginning in 1999.
15. I did this by estimating regressions of the form:
AGE_LT_5i = b0 + b1VAi + b2YEARi + b3(VAi * YEARi) + eI (1)
VAi = 1 if prescription i is a VA prescription
= 0 otherwise
YEARi = the year in which prescription i occurred
If b3< 0, the percentage of new drugs is growing less rapidly (or declining more rapidly) in the VA health system than it is in the rest of the U.S. health-care system.
Estimates of b2, b3, and (b2 + b3) for the three different drug-age measures are shown in the following table:
|b2 + b3||0.000988||0.005809||0.000431
For AGE_LT_5, the VA vs. non-VA difference in the rate of increase of new drug use (b3) is not statistically significant. However, for the other two age measures, the difference is negative and significant.
16. Consider the following econometric model:
AGE_DEATHijt= b POST1990%ijt + aij + dit + gjt + eijt (2)
AGE_DEATHijt= mean age at death from disease i (i = 1,2,…,16) in state j (j = 1,2,…,50) in year t (t=1991,1992,…,2001)
POST1990%ijt= the % of Medicaid prescriptions for disease i in state j in year t that contain active ingredients approved by the FDA after 1990
aij = a fixed effect for disease i in state j
dit = a fixed effect for disease i in year t
gjt = a fixed effect for state j in year t
eijt = a disturbance
The model is to be estimated via weighted least squares, weighting by N_DEATHijt, the number of deaths from disease i in state j in year t.
17. I.e., prescriptions that contain active ingredients approved by the FDA after 1990.
18. These are controlled for by including the gjt’s. The econometric specification is similar to the one that I used in a previous paper, "The Impact of New Drug Launches on Longevity: Evidence from Longitudinal Disease-Level Data from 52 Countries, 1982-2001." In that paper, however, the measure of drug availability was the cumulative number of drugs launched for a given disease in a given country (and the data were subject to left-censoring). The data available for this study are superior in an important respect: we have very extensive data on drugs actually prescribed.
19. See http://www.nber.org/data/deaths.html.
20. See http://www.cms.hhs.gov/medicaid/drugs/drug5.asp.
21. There are about 700 data files: one for each state in each year.
22. I used these data to estimate the following equation:
tot_prod_agecjt = p mdcd_prod_agecjt + acj + dct + gjt + ecjt
tot_prod_agecjt = the mean age (number of years since FDA approval) of all prescriptions in therapeutic class c (c = 1,2,…,6) in region j (j = 1,2,…,55) in month t (t=1,2,…,36)
mdcd_prod_agecjt = the mean age of Medicaid prescriptions in therapeutic class c in region j in month t
acj = a fixed effect for therapeutic class c in region j
dct = a fixed effect for therapeutic class c in year t
gjt = a fixed effect for region j in year t
ecjt = a disturbance
The estimate of p was positive and highly significant (p-value <.0001), which indicates that the extent of use of new drugs in the Medicaid program is strongly correlated with the extent of use of new drugs in general. I will now present the statistics pertaining to b from estimation of eq. (2):
|std. err. ||0.45
The estimate of b is positive and highly significant.
23. b = D * POST1990% = 2.53 * 0.314.
24. = 0.79 / 1.74.
25. Lichtenberg, "The Impact of New Drug Launches on Longevity."
26. See http://www.va.gov/vetdata/demographics/VP2001sn.htm.
27. Since the age groups are five years wide, the probability of surviving from the beginning of age group a to the beginning of age group (a+1) is approximately Sat = (1 – Mat)5. The probability of surviving from the first age group (age < 20) to the beginning of age group a is is Hat = S1t * S2t * … * Sa-1,t. The probability that a person in the first age group will die in age group a is Qat = (Ha+1,t – Hat). The life expectancy of a person in the first age group in year t is is Et = Sa Qat Aa, where Aa= the mean age at death of a person dying in age group a, which I assume to be the midpoint of the age interval. For example, I assume that deaths of people aged 75-79 occur at age 77.5. I assume that people dying after age 100 die at age 102.5.
28. While there are some reasons to expect the mean value of Et to be lower than the mean value of the life expectancy of all U.S. males at birth-serving in the military may impair one's future health-there are other reasons to expect it to be greater. Et is based on a population of individuals who have been veterans, i.e., who lived long enough to serve in the armed forces (e.g., did not die in infancy) and who survived serving in the armed forces. It would be more appropriate to compare Et with the life expectancy of all U.S. males at age twenty, for example. Such data are available for some years (it was 73.25 for 1989-1991 and 75.6 in 2002) but are not available annually (Arias, 2004, Table 11).