Life-expectancy gains from pharmaceutical drugs: a critical appraisal of the literature

Dear Editor,

The article by Paul Grootendorst, Emmanuelle Piérard and Minsup Shim (GPS) provides a useful critique on some of the literature regarding the impact of pharmaceuticals on life expectancy [1]. However, I think that their critique is flawed in certain respects. The most important limitations of their critical appraisal are as follows:

• • Does not clearly distinguish between the effect of pharmaceutical expenditure on life expectancy and the effect of pharmaceutical innovation on life expectancy;

• • Overstates the degree of consensus about the value of pharmaceuticals (or pharmaceutical innovation);

• • Appraisal of the literature is incomplete;

• • Claims that “policy-makers typically are concerned with the impact of individual drugs (or perhaps groups of therapeutically similar drugs) on health outcomes,” but some important pharmaceutical policy questions depend on the impact of drugs in general;

• • Some key findings from previous studies are more robust to their sensitivity analyses than suggested by their conclusions;

• • Misinterprets estimates of the parameters of my model;

• • Does not provide any evidence to support the claim that “analyses of mortality using individual level data or aggregated disease-specific data are more promising,” or acknowledge the limitations of those analyses.

I will now elaborate on these and other aspects of their analysis.

Pharmaceutical innovation versus pharmaceutical expenditure

Most of GPS’s article is devoted to a re-examination of four aggregate econometric studies of life expectancy: Crémieux, Meilleur, Ouellette et al. (CMO) [2]; Frech and Miller (FM) [3]; Shaw, Horrace and Vogel (SHV) [4]; and one of my own studies on this topic [5]. The first three studies attempted to assess the impact of pharmaceutical expenditure on life expectancy. My study was the only study that investigated the impact of pharmaceutical innovation (the launch of new drugs) on life expectancy. The question posed in that study is not whether pharmaceuticals are valuable but whether pharmaceutical innovation is valuable – are new drugs better than old drugs?

I analyzed innovation rather than expenditure because many economists believe that the development of new products is the main reason that people are better-off today than they were several generations ago. Grossman and Helpman argue that “innovative goods are better than older products simply because they provide more ‘product services’ in relation to their cost of production” [6]. Bresnahan and Gordon state simply that “new goods are at the heart of economic progress” [7]. Jones argues that “technological progress [is] the ultimate driving force behind sustained economic growth” (page 2) and that “technological progress is driven by research and development in the advanced world” (page 89) [8]. Bils makes a valid point:

“Much of economic growth occurs through growth in quality as new models of consumer goods replace older, sometimes inferior, models”[101].

Although pharmaceutical expenditure and pharmaceutical innovation may be (positively) correlated – primarily because new drugs are generally more expensive than old drugs – they are far from perfectly correlated. Pharmaceutical expenditure may increase because the number of prescriptions consumed per person increases (e.g., due to aging of the population). I agree with GPS’s argument that failure to control for the age distribution of the population may bias the estimated effect of pharmaceutical expenditure on life expectancy upward owing to reverse causality. An exogenous positive shock to life expectancy would cause an increase in the mean age of the population (holding constant fertility and migration rates), which would cause an increase in per capita pharmaceutical expenditure. Pharmaceutical expenditure may also increase as the average price of existing drugs increases. Ideally, the price indices used to deflate pharmaceutical expenditure would eliminate the effects of such price changes. The price indices used by CMO [2], FM [3] and SHV [4] may not do so, however.

In modern industrialized countries, many thousands of pharmaceutical products (containing ∼2000 active ingredients) are available. In principle, one might hypothesize that longevity depends on the utilization of each of these products or active ingredients. Clearly, even if the detailed data were available, it would not be feasible to estimate a longevity model with thousands of explanatory variables – one for consumption of each product or active ingredient. In order to make inferences about the impact of pharmaceutical consumption on longevity, it is necessary to use a small set of variables to summarize all of this detailed information. Four potential variables are real per capita pharmaceutical expenditure, the number of prescriptions consumed per person, the number of distinct products (or active ingredients) consumed, and the mean vintage (launch year) of prescriptions consumed. The second measure indicates the quantity of drugs consumed. The last may be an indicator of the quality of drugs consumed. In principle, longevity might depend on several characteristics of the full distribution of drugs (e.g., it could depend on the quality, as well as the quantity, of drugs consumed).

If the relative prices of different drugs were equal to their relative benefits (e.g. drug A’s price would be twice as high as drug B’s price if drug A extended life twice as much), then pharmaceutical expenditure would adequately reflect both the quantity and quality of drugs consumed; pharmaceutical expenditure would be a ‘sufficient statistic’. However, in practice, the relative prices of different drugs are unlikely to reflect their relative therapeutic benefits. When a drug loses patent protection in the USA, its (relative) price declines quite dramatically, although its (relative) therapeutic benefits remain unchanged. Therefore, pharmaceutical expenditure is unlikely to be a sufficient statistic – it is unlikely to characterize both the quantity and the quality of all drugs consumed adequately. It is unlikely to be the only indicator of pharmaceutical use that influences health outcomes, and it may not be the best (single) indicator. Lichtenberg found that cardiovascular hospitalization and mortality depended on the quality (vintage) of drugs consumed but not on the quantity [9].

Lack of consensus about the value of pharmaceuticals (or pharmaceutical innovation)

In their article, GPS state:

“There is no question that pharmaceutical drugs are among the most cost effective forms of healthcare, and hold the promise to treat more effectively various debilitating health problems”[1].

Once again, it is important to distinguish between the effects of pharmaceutical use and pharmaceutical innovation on life expectancy. Perhaps most people would agree that substantial reductions in average pharmaceutical use would harm public health, even if they felt there is overuse of certain medications (e.g., owing to marketing activities of pharmaceutical firms).

However, some prominent commentators (innovation deniers [10]) argue that the vast majority of new drugs are no better than older drugs, which implies that a drastic reduction in the number of new drugs introduced would not reduce longevity or harm public health. Angell claims that most of the new drugs launched by drug companies cost far more than older drugs but are no more effective [11]. Light claims:

“Systematic review of actual therapeutic gain in use concludes that, during the past 15 years, only one in seven new drugs offer substantial therapeutic benefit…A 20-year review by industry found that only one in nine new drugs were innovative and therapeutically superior”[12].

Lexchin claims:

“At best, one third of new drugs offer some additional clinical benefit and perhaps as few as 3% are major therapeutic advances”[102].

The proposition that, overall, pharmaceutical innovation has made major contributions to longevity growth (let alone that it has been cost effective) is not universally accepted, and is worthy of empirical verification.

Policy relevance of the impact of drugs in general

Grootendorst, Piérard and Shim claim “policy-makers typically are concerned with the impact of individual drugs (or perhaps groups of therapeutically similar drugs) on health outcomes” but some important pharmaceutical policy questions depend on the impact of drugs in general. An important current issue is the optimal duration of market exclusivity for biologic drugs [103]. A bitter congressional fight over the cost of very expensive biotechnology drugs now focuses on a single, hotly debated topic: how many years should makers of those drugs be exempt from generic competition? Policy-makers want to allow competition without undermining the financial incentives the pharmaceutical industry needs to undertake the risky job of developing the next drugs for cancer and other diseases, which is where the issue of the ‘magic’ year duration comes in. Trade groups for the big pharmaceutical and biotechnology companies say that to recoup their investments, they need an exclusivity period free of generic competition that would last 12–14 years from the time the US FDA approves a drug for sale. However, consumer groups, insurers, employers and generic drug companies state anything more than 5 years would eviscerate any potential savings from the new competition.

The theoretically optimal duration of the exclusivity period depends on the magnitude of the health benefit (e.g., gain in life expectancy) from the average new biotech drug (among other things), not on the gain from a specific drug.

Incomplete appraisal of the literature

The title of GPS’s article – “Life-expectancy gains from pharmaceutical drugs: a critical appraisal of the literature” – suggests that their article will appraise all, or almost all, of the studies that have been published on this topic. However, numerous studies of the impact of pharmaceutical innovation on longevity are not mentioned. Many of these use individual-level or disease-specific data, which GPS suggest may yield more reliable estimates than aggregate data.

Lichtenberg [104], Jung et al.[13], and Lichtenberg et al.[14] analyzed the effect of pharmaceutical innovation on mortality using patient-level data from Puerto Rico, South Korea and Quebec, respectively. The measure of pharmaceutical innovation they used was the mean vintage of prescription drugs used by the patient. The vintage of a good is the year in which the good was first used. For example, the vintage of the drug atorvastatin (Lipitor^®) is 1997 – the year that the drug was approved by the FDA. Those studies sought to test the hypothesis that, ceteris paribus, people using newer or later vintage drugs will be in better health and, therefore, will live longer.

This hypothesis is predicated on the idea that these goods and services, similar to other research-and-development-intensive products, are characterized by embodied technological progress. Solow argued:

“Many if not most innovations need to be embodied in new kinds of durable equipment before they can be made effective. Improvements in technology affect output only to the extent that they are carried into practice either by net capital formation or by the replacement of old-fashioned equipment by the latest models” (page 91) [15].

Innovations may be embodied in nondurable goods (e.g., drugs) and services, as well as in durable equipment.

A number of econometric studies have investigated the hypothesis that capital equipment employed by US manufacturing firms embodies technological change, that is, that each successive vintage of investment is more productive than the last [16–19]. Equipment is expected to embody significant technical progress owing to the relatively high research and development intensity of equipment manufacturers. The method that has been used to test the equipment-embodied technical-change hypothesis is to estimate manufacturing production functions, including (mean) vintage of equipment, as well as quantities of capital and labor. These studies have concluded that technical progress embodied in equipment is a major source of manufacturing productivity growth.

Although most previous empirical studies of embodied technical progress have focused on the equipment used in manufacturing, embodied technical progress may also be an important source of progress in healthcare. One important input in the production of health is pharmaceuticals. According to the National Science Foundation, the research and development intensity of drug and medicine manufacturing is 74% higher than the research and development intensity of machinery and equipment manufacturing. Therefore, it is quite plausible that there is also a high rate of pharmaceutical-embodied technical progress. Lichtenberg [104], Jung et al.[13], and Lichtenberg et al.[14] all found that patients using newer drugs had lower mortality rates when controlling for age, sex, region, diagnosis, number of diagnoses and utilization of medical services.

Other studies have used aggregate, disease-specific data (i.e., data on pharmaceutical use by, and mortality of, all people with a given disease in a given region and year). This approach has been applied to different sets of diseases and countries, using different measures of pharmaceutical innovation.

A number of studies have examined data on the entire cross-section of diseases. In some of these studies, the measure of pharmaceutical innovation was the cumulative number of drugs approved to treat a given disease. Lichtenberg examined the effect of changes in the cumulative number of drugs approved to treat a given disease on mortality from that disease using longitudinal data on all diseases in the USA [20,21]. In addition, Lichtenberg examined this effect using longitudinal disease-level data from 52 developed and developing countries during the period 1982–2001 [22]. All three studies found that the introduction of new drugs has made important contributions to the growth in life expectancy.

In other studies of the entire cross-section of diseases, the measure of pharmaceutical innovation was the mean vintage of drugs used to treat the disease. As GPS observe, “new molecular entity approvals do not affect life expectancy directly; they affect life expectancy only to the extent that they are consumed.” Since mean vintage is based on data on drugs actually used, it may be a better measure of pharmaceutical innovation than the cumulative number of drugs approved. Lichtenberg and Duflos [23] performed a similar study to Lichtenberg’s USA study [24], using longitudinal data on all diseases in Australia.

Other studies have examined narrower (but still rather broad) sets of diseases. Lichtenberg examined the impact of chemotherapy innovation and other factors on the survival of US cancer patients during the period 1992–2003 [105]. Of particular focus was whether cancer survival rates increased more for those cancer sites that had the largest increases in the proportion of chemotherapy treatments that were ‘new’ treatments (i.e., the largest increases in chemotherapy vintage). Other types of medical innovation (i.e., other pharmaceutical innovation and innovation in surgical procedures, diagnostic radiology procedures and radiation oncology procedures) were controlled for. It was found that the cancer sites where chemotherapy vintage (measured by the share of post-1990 treatments) increased the most during the period 1992–2003 tended to have larger increases in observed survival rates, ceteris paribus.

Lichtenberg also examined the impact of chemotherapy innovation on cancer survival but over a longer period (1978–2004), and measured innovation by changes in the number of drugs that had been introduced to treat that cancer [25]. Variables likely to reflect changes in diagnostic techniques (e.g., cancer stage distribution, age at diagnosis, number of people diagnosed [incidence], and use of surgery and radiation) were controlled for. New cancer drugs introduced during the period 1968–1994 were estimated to have increased the life expectancy of people diagnosed with cancer by almost 1 year (from ∼9 to 10 years). Since the lifetime risk of being diagnosed with cancer is approximately 40%, the 1978–2004 increase in the lagged stock of cancer drugs is estimated to have increased the life expectancy of the entire US population by 0.38 years. This represents approximately 9% of the overall increase in US life expectancy at birth.

Lichtenberg and Waldfogel examined the impact of pharmaceutical innovation on mortality from a different set of diseases – orphan diseases [106]. The number of drugs available to treat these diseases increased dramatically after the 1983 Orphan Drug Act, which provided important financial incentives for the development of drugs for the treatment of rare conditions. Lichtenberg and Waldfogel showed that utilization of drugs by people with rare diseases increased, relative to utilization of drugs by people with common conditions, after the Orphan Drug Act [106]. Moreover, the relative mortality rates of people with rare conditions decreased.

Some key findings from previous studies are more robust to their sensitivity analyses than suggested by GPS’s conclusions

Grootendorst, Piérard and Shim state that “the models that we replicated were found to be sensitive to seemingly innocuous changes in specification.” I have two comments about their ‘replication’ of my model. First, their change in specification is not necessarily appropriate. Second, some of my key findings are robust to their change in specification.

Grootendorst, Piérard and Shim replaced lagged real per capita publicly funded healthcare spending with lagged real per capita total spending on hospitals, physicians and public health.

“The healthcare spending variable [Lichtenberg] uses should reflect nonpharmaceutical healthcare spending since drug spending would not appear in a reduced form model.”

I disagree with the notion that only a single indicator relevant to pharmaceuticals should be included in the life-expectancy model. As noted previuosly, pharmaceutical utilization is a highly complex phenomenon involving thousands of products, and there is good reason to doubt that a single indicator would capture all longevity-relevant information regarding pharmaceutical consumption. Production functions that allow for capital-embodied technological progress include both the quantity and vintage (quality) of capital. Similarly, according to Romer’s model of endogenous technical change, output (health outcomes) depends on the ‘stock of ideas’, as well as on the quantities of inputs. The number of approved drugs is analogous to the stock of ideas [26].

Although GPSs estimate of the impact of new molecular entities approvals on life expectancy is quite different from my estimate, GPS confirm two of my key findings. First, new molecular entity approvals have had a highly significant positive impact on life expectancy (which has also been confirmed by Schnittker [27]). Second, the per capita expenditure on drug development required to increase life expectancy at birth by 1 year is a very small fraction of consensus estimates of the value of a life-year.

Misinterpretation of parameter estimates

I believe that GPS misinterpret the estimates of both the original and ‘replicated’ versions of my model. Contrary to their statement, I do not assume that “only newborns would benefit from increased healthcare spending.” Presumably, GPS claim this because my calculation of the cost per life-year gained from increased public health expenditure (and from increased new molecular entities) is based on the average number of births per year, rather than average population. However, it would be inappropriate to base this calculation on average population. Since the dependent variable is life expectancy at birth, the coefficient on public health expenditure measures the effect of this variable on mortality over the entire lifecycle (i.e., the product of the effects on mortality at every age). To impute all of this longevity increase to each person alive every year would grossly overestimate the longevity benefits from increased public health expenditure or increased new molecular entities. The annual longevity benefit is the reduction in mortality during the next year of life, not over the entire lifecycle.

Alternative approaches to estimating the impact of pharmaceutical innovation

Grootendorst, Piérard and Shim conclude that “analyses using individual level data or perhaps disease specific data will probably produce more compelling results” than analyses based on aggregate data. However, since they have only performed sensitivity analyses of studies based on aggregate data, they have not established the superiority of studies based on patient-level or disease-specific data. Their conclusion is really a conjecture.

I concur with the view that studies based on individual-level or disease-specific data can provide valuable evidence regarding the impact of pharmaceutical innovation on longevity. In addition, GPS’s conjecture may well be correct: these studies may provide the most reliable evidence. However, in my view, each study design has both advantages and disadvantages, and each has the potential to increase our understanding of this issue.

Individual-level studies that involve large numbers of drugs (hence enabling inferences about the general impact of pharmaceutical innovation) are unlikely to be based on samples in which drugs are randomly assigned to patients. Moreover, data on potentially confounding variables (e.g., education, race and income) are frequently unavailable at the patient level, and the follow-up period may be short.

Some of these problems can be overcome by judicious use of aggregated, disease-specific data (in part, because data from a variety of sources can be linked). However, there is an important obstacle to making accurate inferences about the impact of pharmaceutical innovation on longevity from disease-specific data. Use of newer drugs may have cross-disease spillover effects: using newer drugs for one disease may either increase or decrease mortality from other diseases (in part, owing to ‘competing risks’). Such spillovers could be negative or positive. For example, using a newer drug to treat cardiovascular disease might reduce cardiovascular disease mortality but increase life-years lost due to cancer. On the other hand, using a newer drug to treat depression and other mental disorders might lead to better management of cardiovascular disease.

Yours faithfully,