The impact of pharmaceutical innovation on the longevity and hospitalization of New Zealand cancer patients, 1998–2017

ABSTRACT

Background

We investigate whether the cancer sites that experienced more pharmaceutical innovation in New Zealand had larger subsequent declines in premature mortality and hospitalization rates and larger subsequent increases in 5-year survival rates, controlling for changes in incidence. 

Research design and methods

We estimate the effects of the number of WHO ATC5 chemical substances and ATC4 chemical subgroups approved on the number of years of potential life lost before ages 85, 75, 65, 5-year relative survival rates, and the number of inpatient hospital discharges, by estimating difference-in-differences (2-way fixed-effects) models using aggregate longitudinal data on 23 cancer sites.

Results

Substances/subgroups approved during 1985-2001 reduced the number of years of potential life lost before age 85 (YPLL85) in 2017 by 67%. Those substances/subgroups reduced YPLL75 and YPLL65 in 2017 by similar percentages. The odds of surviving at least 5 years after diagnosis are significantly positively related to the number of substances previously approved.

Conclusions

The cost per life-year before age 85 gained in 2017 from previous drug approvals did not exceed 1719 USD. The WHO considers interventions whose cost per quality-adjusted life-year gained is less than per capita GDP to be highly cost-effective; New Zealand’s per capita GDP in 2017 was 42,260 USD. capita GDP in 2017 was 42,260 USD.

Expert Opinion

Pharmaceutical innovation—the introduction and use of new drugs—substantially increased cancer survival rates in New Zealand, and substantially reduced premature (before ages 85, 75, and 65) cancer mortality there during the period 1998–2017. Moreover, overall the new cancer drugs were highly costeffective. The drugs approved during 1985–2001 are estimated to have reduced the number of years of potential life lost before age 85 in 2017 by 244,876. Even if previous drug approvals increased the cost of hospital discharges and other medical costs, the cost per life-year before age 85 gained in 2017 from those approvals could not have exceeded 1719 USD.

KEYWORDS:

• Pharmaceutical
• innovation
• longevity
• hospitalization
• New Zealand
• cancer

1. Introduction

A variety of statistics indicate that the longevity of New Zealand cancer patients has increased since the year 2000 or thereabouts. Between 1998 and 2010, the 5-year survival rate for all adult cancers increased from 57.7% to 63.3%, and the 5-year survival rate for all childhood (ages 0–14) cancers increased from 71.6% to 84.5% [1]. In principle, these increases could overstate true survival gains due to earlier diagnosis and increasing lead-time bias.¹ But other indicators that are not affected by changing patterns of diagnosis also point to longevity gains. Between 2001 and 2015, the premature (before age 75) cancer mortality rate (the number of potential years of life lost due to malignant neoplasms before age 75 per 100,000 people below age 75) declined 22.3% (from 1689.2 to 1312.6) [2] In principle, this could be attributable to declining cancer incidence. The age-standardized cancer incidence rate declined between 2001 and 2015, but by one-third as much (7.0%, from 359.8 to 334.6) as the premature mortality rate [3].

Although the longevity of New Zealand cancer patients has increased overall, there has been considerable variation across cancer sites (e.g. breast, prostate, colon) in the size of the increase. For example, the 1998–2010 increase in the 5-year survival rate for prostate cancer (from 81.9% to 91.1%) was much greater than the increase for Hodgkin lymphoma (from 83.0% to 84.5%). Figure 1 shows the percentage change between 2001 and 2015 in the premature (before age 75) mortality rate for 12 cancers. Five of the 12 cancer sites had declines in the premature mortality rate of at least 30%, but three sites had increases of at least 10%.

Figure 1. Change in premature (before age 75) mortality rate, 2001–2015

In this paper, we will analyze the effect that pharmaceutical innovation – the introduction and use of new drugs used to treat cancer – had on the longevity and hospitalization of New Zealand cancer patients during the period 1998–2017. Between 1979 and 1999, the number of drugs (ATC5 chemical substances) used to treat cancer ever approved in New Zealand increased by 66%, from 74 to 123; between 1999 and 2019, the number increased by 53%, to 188.

The analysis will be performed using a difference-in-differences research design based on aggregate data – longitudinal data on the 23 cancer sites defined in the New Zealand Ministry of Health historical summary [3]. In essence, we will investigate whether the cancer sites that experienced more pharmaceutical innovation had larger subsequent declines in premature mortality and hospitalization rates and larger subsequent increases in 5-year survival rates, controlling for changes in incidence. As shown in Figure 2, there has been considerable variation across cancer sites in the number of new drugs. During the period 1994–2019, 17 new drugs for treating cervical cancer were approved, while 25 new drugs for treating colorectal cancer were approved. A complete list of drugs used to treat cancer (according to the Therapeutic Target Database 2020 [4]) and approved by the New Zealand Medicines and Medical Devices Safety Authority (MEDSAFE) [5] and their indications is shown in Appendix Table 1.

Figure 2. Number of chemical substances used to treat 4 cancer sites ever approved by Medsafe, 1994‐2019

Table 1. Weights used for each of the outcome measures

Outcome Measure	Weight
YPLL85s,t	Σt YPLL85s,t
YPLL75s,t	Σt YPLL75s,t
YPLL65s,t	Σt YPLL65s,t
SURV5%s,t/(1 – SURV5%s,t)	N_CASESs,t
INPATIENTs,t	Σt INPATIENTs,t

In Section 2, we describe an econometric model of cancer patient outcomes. The data sources used to construct the data to estimate this model are described in Section 3. Empirical results are presented in Section 4. Key implications of the estimates are discussed in Section 5. Section 6 provides a summary and conclusions.

2. Cancer patient outcomes model

In his model of endogenous technological change, Romer [6] hypothesized an aggregate production function such that an economy’s output depends on the ‘stock of ideas’ that have previously been developed, as well as on the economy’s endowments of labor and capital. The model that we will estimate may be considered a health production function, in which the outcome is an indicator of health output, and the cumulative number of drugs approved is analogous to the stock of ideas. New drug approvals can improve outcomes for two reasons. First, the quality of newer products may be higher than the quality of older products, as in ‘quality ladder’ models [7]. Second, ‘one of the principal means, if not the principal means, through which countries benefit from international trade is by the expansion of varieties’ [8].

The model will be of the following form:

(1) $ln\left(OUTCOM{E}_{s,t}\right)={\beta }_{k}CUM\mathrm{\_}DRUG{S}_{s,t-k}+{\pi }_{k}CUM\mathrm{\_}GROUP{S}_{s,t-k}+\gamma ln\left(INCIDENC{E}_{s,t}\right)+{\alpha }_s+{\delta }_t+{\epsilon }_{s,t}$(1)

where OUTCOME_s,t is one of the following variables:

YPLL85s,t	=	= the number of years of potential life lost before age 85 due to cancer at site s in year t (t = 2000, 2005, 2009, 2013, 2017)
YPLL75s,t	=	= the number of years of potential life lost before age 75 due to cancer at site s in year t
YPLL65s,t	=	= the number of years of potential life lost before age 65 due to cancer at site s in year t
SURV5%s,t/(1 – SURV5%s,t)	=	= the odds of surviving at least 5 years after diagnosis with cancer at site s in year t2 (t = 1998, 2000, 2002, 2004, 2006, 2008, 2010)
INPATIENTs,t	=	= the number of publicly-funded3 inpatient hospital discharges for cancer at site s in year t (t = 2005, 2012, 2017)

and

CUM_DRUGSs,t-k	=	= ∑d INDds APPROVEDd,t-k = the number of WHO ATC5 chemical substances (drugs) to treat cancer at site s that had been approved in New Zealand by the end of year t-k (k = 0, 2, 4, …, 24)
INDds	=	= 1 if drug d is used to treat (indicated for) cancer at site s, according to the Therapeutic Target Database in October 2020 = 0 if drug d is not used to treat (indicated for) cancer at sites
APPROVEDd,t-k	=	= 1 if drug d was approved in New Zealand by the end of year t-k = 0 if drug d was not approved in New Zealand by the end of year t-k
CUM_GROUPSs,t-k	=	= ∑g INDgs APPROVEDg,t-k = the number of WHO ATC4 chemical subgroups to treat cancer at site s that had been approved in New Zealand by the end of year t-k (k = 0, 2, 4, …, 24)
INDgs	=	= 1 if any drug in ATC4 chemical subgroup g is used to treat (indicated for) cancer at site s, according to the Therapeutic Target Database in October 2020 = 0 if no drug in ATC4 chemical subgroup g is used to treat (indicated for) cancer at site s
APPROVEDg,t-k	=	= 1 if any drug in ATC4 chemical subgroup g was approved in New Zealand by the end of year t-k = 0 if no drug in ATC4 chemical subgroup g was not approved in New Zealand by the end of year t-k
INCIDENCEs,t	=	= the average annual number of patients diagnosed with cancer at site s in years t-4 to year t
αs	=	= a fixed effect for cancer at site s
δt	=	= a fixed effect for year t

The first three outcome measures (YPLL85, YPLL75, and YPLL65) are indicators of the burden of cancer. These measures are not conditional on diagnosis and therefore are not subject to lead-time bias. Years of potential life lost is the weighted sum of the number of deaths from a specific disease or cause, weighted by the number of years prior to an age threshold (e.g. age 85) at which the deaths occurred.⁴ Data on cause-specific YPLL are provided in many online databases, including the U.S. CDC’s WISQARS Years of Potential Life Lost (YPLL) Report [9], Eurostat [10], the OECD Health Statistics 2020 database [11], and the WHO Global Health Estimates database [12]. The World Health Organization has used YPLL to measure disease burden in its Global Burden of Disease (GBD) and Global Health Estimates (GHE) reports for many years. In the 2010 GBD, the WHO used an age threshold of 86.01 years for all persons. In the current GHE, the WHO uses an age threshold of 91.93 years for all persons [13]. The U.S. CDC’s WISQARS Years of Potential Life Lost (YPLL) Report website [9] permits one to specify age thresholds of 65, 70, 75, 80, and 85.

Chemical substances are divided into different groups according to the organ or system on which they act and their therapeutic, pharmacological, and chemical properties. In the Anatomical Therapeutic Chemical (ATC) classification system developed by the World Health Organization Collaborating Centre for Drug Statistics Methodology, drugs are classified in groups at five different levels. The highest (1st) level is the ‘anatomical main group’ level; there are 14 anatomical main groups. The 2nd, 3rd, 4th, and 5th levels are ‘therapeutic subgroup,’ ‘pharmacological subgroup,’ ‘chemical subgroup,’ and ‘chemical substance,’ respectively.⁵ The effect on outcomes of the approval in year t-k of the first chemical substance in a chemical subgroup may differ from the effect of a subsequent chemical substance in a chemical subgroup. For example, the approval in 1981 of the first substance in the platinum compound subgroup (L01XA), cisplatin (L01XA01) may have had a different effect than the approval in 1987 of another platinum compound, carboplatin (L01XA02). Eq. (1)(1) $ln\left(OUTCOM{E}_{s,t}\right)={\beta }_{k}CUM\mathrm{\_}DRUG{S}_{s,t-k}+{\pi }_{k}CUM\mathrm{\_}GROUP{S}_{s,t-k}+\gamma ln\left(INCIDENC{E}_{s,t}\right)+{\alpha }_s+{\delta }_t+{\epsilon }_{s,t}$(1) allows the effects to differ: the effect of the first chemical substance in a chemical subgroup is (β_k + π_k); the effect of a subsequent chemical substance in a chemical subgroup is β_k. We will estimate three versions of eq. (1)(1) $ln\left(OUTCOM{E}_{s,t}\right)={\beta }_{k}CUM\mathrm{\_}DRUG{S}_{s,t-k}+{\pi }_{k}CUM\mathrm{\_}GROUP{S}_{s,t-k}+\gamma ln\left(INCIDENC{E}_{s,t}\right)+{\alpha }_s+{\delta }_t+{\epsilon }_{s,t}$(1) : one that includes the number of substances but not the number of subgroups; one that includes the number of subgroups but not the number of substances; and one that includes both variables. Since the two variables are highly correlated, in the third version it will be difficult to isolate the effect of either variable, but their joint effect can be accurately estimated.

In principle, it might be desirable to disaggregate substances in other dimensions, e.g. to distinguish between therapeutic and supportive agents, or between biologic/targeted agents and chemotherapy agents, although cross-classifying substances in multiple dimensions are likely to yield imprecise estimates. I am not aware of any publicly available databases (the only type of data relied upon in this article) that would enable us to make those distinctions. (For example, the FDA’s Drugs Approved for Different Types of Cancer website [14] does not make those distinctions.) Clinicians may be able to systematically distinguish between different types of agents, and this may be a fruitful area of future research. In the meantime, our estimates may capture the average effect of different types of agents.

Inclusion of year and cancer-site fixed effects controls for the overall change in outcomes⁶ and for stable between-disease differences in outcomes. When OUTCOME_s,t = YPLL85_s,t, a negative and significant estimate of β_k in eq. (1)(1) $ln\left(OUTCOM{E}_{s,t}\right)={\beta }_{k}CUM\mathrm{\_}DRUG{S}_{s,t-k}+{\pi }_{k}CUM\mathrm{\_}GROUP{S}_{s,t-k}+\gamma ln\left(INCIDENC{E}_{s,t}\right)+{\alpha }_s+{\delta }_t+{\epsilon }_{s,t}$(1) would signify that cancer sites for which there were more drug approvals had larger subsequent declines in premature (before age 85) mortality. When OUTCOME_s,t = SURV5%_s,t/(1 – SURV5%_s,t), a positive and significant estimate of β_k in eq. (1)(1) $ln\left(OUTCOM{E}_{s,t}\right)={\beta }_{k}CUM\mathrm{\_}DRUG{S}_{s,t-k}+{\pi }_{k}CUM\mathrm{\_}GROUP{S}_{s,t-k}+\gamma ln\left(INCIDENC{E}_{s,t}\right)+{\alpha }_s+{\delta }_t+{\epsilon }_{s,t}$(1) would signify that cancer sites for which there were more drug approvals had larger subsequent increases in the 5-year survival rate.

The standard errors of eq. (1)(1) $ln\left(OUTCOM{E}_{s,t}\right)={\beta }_{k}CUM\mathrm{\_}DRUG{S}_{s,t-k}+{\pi }_{k}CUM\mathrm{\_}GROUP{S}_{s,t-k}+\gamma ln\left(INCIDENC{E}_{s,t}\right)+{\alpha }_s+{\delta }_t+{\epsilon }_{s,t}$(1) will be clustered within cancer sites. The data exhibit heteroscedasticity. For example, cancer sites with larger mean premature mortality during 2000–2017 had smaller (positive and negative) annual percentage fluctuations in YPLL85_s,t. Eq. (1)(1) $ln\left(OUTCOM{E}_{s,t}\right)={\beta }_{k}CUM\mathrm{\_}DRUG{S}_{s,t-k}+{\pi }_{k}CUM\mathrm{\_}GROUP{S}_{s,t-k}+\gamma ln\left(INCIDENC{E}_{s,t}\right)+{\alpha }_s+{\delta }_t+{\epsilon }_{s,t}$(1) will therefore be estimated by weighted least-squares. The weights used for each of the outcome measures are shown in (Table 1).⁷

Estimates of the parameters of eq. (1)(1) $ln\left(OUTCOM{E}_{s,t}\right)={\beta }_{k}CUM\mathrm{\_}DRUG{S}_{s,t-k}+{\pi }_{k}CUM\mathrm{\_}GROUP{S}_{s,t-k}+\gamma ln\left(INCIDENC{E}_{s,t}\right)+{\alpha }_s+{\delta }_t+{\epsilon }_{s,t}$(1) will enable us to estimate the contribution of previous drug approvals to changes in outcomes. For example, the contribution of drug approvals during the years 2000-k + 1 to 2017-k to the 2000–2017 change in outcomes are ϕ_k = [β_k * mean(ΔCUM_DRUGS_k) + π_k * mean(ΔCUM_GROUPS_k)], where, for example, mean(ΔCUM_DRUGS_k) = mean(CUM_DRUGS_s,2017-k – CUM_DRUGS_s,2000-k), and mean() is the (weighted) mean across cancer sites.⁸ The percentage change in the outcome in 2017 due to the new drugs approved during the years 2000-k + 1 to 2017-k is exp(ϕ_k) – 1.

Due to data limitations, the number of new chemical entities and incidence are the only disease-specific, time-varying, explanatory variables in eq. (1)(1) $ln\left(OUTCOM{E}_{s,t}\right)={\beta }_{k}CUM\mathrm{\_}DRUG{S}_{s,t-k}+{\pi }_{k}CUM\mathrm{\_}GROUP{S}_{s,t-k}+\gamma ln\left(INCIDENC{E}_{s,t}\right)+{\alpha }_s+{\delta }_t+{\epsilon }_{s,t}$(1) . But both a patient-level U.S. study and a longitudinal country-level study have shown that controlling for numerous other potential determinants of longevity does not reduce, and may even increase, the estimated effect of pharmaceutical innovation. The study based on patient-level data [15] found that controlling for race, education, family income, insurance coverage, Census region, BMI, smoking, the mean year the person started taking his or her medications, and over 100 medical conditions had virtually no effect on the estimate of the effect of pharmaceutical innovation (the change in drug vintage) on life expectancy. The study based on longitudinal country-level data [16] found that controlling for ten other potential determinants of longevity change (real per capita income, the unemployment rate, mean years of schooling, the urbanization rate, real per capita health expenditure (public and private), the DPT immunization rate among children ages 12–23 months, HIV prevalence and tuberculosis incidence) increased the coefficient on pharmaceutical innovation by about 32%.

Failure to control for non-pharmaceutical medical innovation (e.g. innovation in diagnostic imaging, surgical procedures, and medical devices) is also unlikely to bias estimates of the effect of pharmaceutical innovation on premature mortality, for two reasons. First, according to Dorsey et al. [17], 88% of private U.S. funding for biomedical research came from pharmaceutical and biotechnology firms. Government funding has also played an important role in pharmaceutical innovation. The National Cancer Institute [18] says that it ‘has played a vital role in cancer drug discovery and development, and, today, that role continues.’ Second, previous research based on U.S. data [19,20] indicates that non-pharmaceutical medical innovation is not positively correlated across diseases with pharmaceutical innovation.

While increased cancer screening may reduce cancer mortality, changes in screening intensity may not be correlated across cancer sites with the number of new drug approvals. Also, some studies have found no mortality benefit from more intensive screening. For example, data from the Prostate, Lung, Colorectal, and Ovarian randomized screening trial showed that, after 13 years of follow up, men who underwent annual prostate cancer screening with prostate-specific antigen testing and digital rectal examination had a 12% higher incidence of prostate cancer than men in the control group but the same rate of death from the disease. No evidence of a mortality benefit was seen in subgroups defined by age, the presence of other illnesses, or pre-trial PSA testing [21].

The measure of pharmaceutical innovation in eq. (1(1) $ln\left(OUTCOM{E}_{s,t}\right)={\beta }_{k}CUM\mathrm{\_}DRUG{S}_{s,t-k}+{\pi }_{k}CUM\mathrm{\_}GROUP{S}_{s,t-k}+\gamma ln\left(INCIDENC{E}_{s,t}\right)+{\alpha }_s+{\delta }_t+{\epsilon }_{s,t}$(1) ) – the number of chemical substances previously approved to treat a disease – is not the theoretically ideal measure. Outcomes are presumably more strongly related to the drugs actually used to treat a disease than it is to the drugs that could be used to treat the disease. A preferable measure is the mean vintage of drugs used to treat cancer at site s in year t, defined as VINTAGE_st = ∑_d Q_dst LAUNCH_YEAR_d/∑_d Q_dst, where Q_dst = the quantity of drug d used to treat cancer at site s in year t, and LAUNCH_YEAR_d = the world launch year of drug d.⁹ Unfortunately, measurement of VINTAGE_st is infeasible: even though data on the total quantity of each drug sold in each year (Q_d.t = Σ_s Q_dst) are available, many drugs are used to treat multiple diseases, and from the data available to us it was not possible to determine the quantity of drug d used to treat cancer at site s in year t.¹⁰ However, Lichtenberg [19] showed that, in France during the period 2000–2009, there was a highly significant positive correlation across drug classes between changes in the (quantity-weighted) vintage of drugs and changes in the number of chemical substances previously commercialized within the drug class.

In eq. (1(1) $ln\left(OUTCOM{E}_{s,t}\right)={\beta }_{k}CUM\mathrm{\_}DRUG{S}_{s,t-k}+{\pi }_{k}CUM\mathrm{\_}GROUP{S}_{s,t-k}+\gamma ln\left(INCIDENC{E}_{s,t}\right)+{\alpha }_s+{\delta }_t+{\epsilon }_{s,t}$(1) ), premature mortality from cancer at site s in year t depends on the number of new chemical entities (drugs) to treat cancer at site s approved in New Zealand by the end of year t-k, i.e. there is a lag of k years. Eq. (1)(1) $ln\left(OUTCOM{E}_{s,t}\right)={\beta }_{k}CUM\mathrm{\_}DRUG{S}_{s,t-k}+{\pi }_{k}CUM\mathrm{\_}GROUP{S}_{s,t-k}+\gamma ln\left(INCIDENC{E}_{s,t}\right)+{\alpha }_s+{\delta }_t+{\epsilon }_{s,t}$(1) will be estimated for different values of k: k = 0, 2, 4, … , 24.¹¹ One would expect there to be a substantial lag, for two reasons.

The first reason is that new drugs diffuse gradually – they will not be used widely until years after approval. Figure 3 shows data on the mean annual number of standard units¹² of cancer drugs sold in New Zealand during 2007–2017, by age of drug, i.e. by the number of years since the drug was approved in New Zealand. Mean utilization of drugs that are 12 years old is about 4 times as great as mean utilization of drugs that are 6 years old; mean utilization of drugs that are 18 years old is 3.7 times as great as mean utilization of drugs that are 12 years old. The relatively low utilization of new drugs may be due to several factors. One is that the prices of old drugs (most of which are no longer patent-protected) are considerably lower than the prices of new, patent-protected drugs. Moreover, the entire cost of very new drugs is borne entirely by patients: Barber and Sheehy [22] noted that the mean lag between regulatory approval of a drug in New Zealand and its inclusion in the New Zealand Pharmaceutical Schedule (a list of the prescription medicines and therapeutic products subsidized by the Government) is 23.7 months. A second factor may be that it takes time for physicians to become knowledgeable about new treatment options. A third potential factor is that new drugs may be targeted at smaller patient populations. Data from the U.S. Food and Drug Administration [23] indicate that drugs approved by the FDA since 2000 were twice as likely to include pharmacogenomic information in their labeling as drugs approved before 2000. A fourth potential factor is that older drugs are more likely to have supplemental indications, i.e. indications approved after the drug was initially approved, than new drugs.

Figure 3. Mean annual number (in millions) of standard units of drugs used to treat cancer sold in New Zealand during 2007‐2017, by number of years since approval

The second reason for a long lag from drug approval to mortality reduction is that there is usually a substantial lag from diagnosis (when drug treatment is likely to begin and be most intensive) to death. The 5-year observed survival rate of all adult patients diagnosed with cancer in 1998 was 57.7%.

The effect of a drug’s approval on premature mortality is likely to depend on both the quality and the quantity of the drug. Indeed, it is likely to depend on the interaction between quality and quantity: a quality improvement will have a greater impact on mortality if drug utilization (quantity) is high. Although newer drugs tend to be of higher quality than older drugs [24], the utilization of very new drugs is quite low, so the impact on mortality of very new drugs may be lower than the impact of older drugs.

3. Data

Mortality data (YPLL85, YPLL75, YPLL65). Data on the number of years of potential life lost before ages 85, 75, and 65, by cancer site and year (2000, 2005, 2009, 2013), were constructed from data contained in the WHO’s Cause of Death Query online [25].¹³ Data for 2017 were constructed from the New Zealand Ministry of Health’s Mortality 2017 data tables [26]. These datasets provide data on deaths by underlying cause and age. Underlying cause of death is defined as ‘the disease or injury which initiated the train of morbid events leading directly to death, or the circumstances of the accident or violence which produced the fatal injury’ in accordance with the rules of the International Classification of Diseases.

Cancer survival data (SURV5%). Data on 5-year relative survival rates, by cancer site and year (1998–2010), were obtained from the Ministry of Health [1].

Publicly funded inpatient hospital discharges (INPATIENT). Data on the number of publicly funded inpatient hospital discharges in 2004–2005, 2012–2013, and 2016–2017 were obtained from the Ministry of Health [27–29].

Cancer incidence data (INCIDENCE). Data on the number of new cancer cases, by cancer site and year (1948–2016), were obtained from the Ministry of Health [3].

NCE approvals in New Zealand (APPROVED). We obtained data on the earliest dates at which chemical substances were approved by the New Zealand Medicines and Medical Devices Safety Authority, MEDSAFE, from that Authority’s website [5].

Drug indications (IND). Data on drug indications were obtained from the ‘drug to disease mapping with ICD identifiers’ file of the Therapeutic Target Database [4,30]. Pharmac (the New Zealand government agency that decides which pharmaceuticals to publicly fund in New Zealand) may not subsidize a drug for all of the indications listed in the Therapeutic Target Database. For instance, a medicine might be indicated for use against both breast and lung cancer in the Therapeutic Target Database but is only subsidized for use against breast cancer in New Zealand. Some patients may pay out of pocket to use it for a clinical indication different from the one funded by Pharmac, but its overall utilization, and impact on health outcomes, would be lower than if it were available at a subsidized price for both sites.

Unit sales of cancer drugs. Data on total market unit sales of cancer drugs in New Zealand, by WHO ATC5 chemical substance and year (2007–2017), were obtained from the IQVIA MIDAS database.

4. Empirical results

Now we will present estimates of eq. (1)(1) $ln\left(OUTCOM{E}_{s,t}\right)={\beta }_{k}CUM\mathrm{\_}DRUG{S}_{s,t-k}+{\pi }_{k}CUM\mathrm{\_}GROUP{S}_{s,t-k}+\gamma ln\left(INCIDENC{E}_{s,t}\right)+{\alpha }_s+{\delta }_t+{\epsilon }_{s,t}$(1) . All estimated models included ln(INCIDENCE_s,t) (the log of the average annual number of patients diagnosed with cancer at site s in years t-4 to year t), cancer site fixed effects, and year fixed effects. The coefficient on the incidence measure (γ) was positive and significant in all the premature mortality models. Estimates of γ were close to 0.6: a 10% increase in the number of new cases in years t-4 to year t was associated with about a 6% increase in the number of years of life lost in year t. To conserve space, we will report only estimates of β_k.

Estimates of eq. (1)(1) $ln\left(OUTCOM{E}_{s,t}\right)={\beta }_{k}CUM\mathrm{\_}DRUG{S}_{s,t-k}+{\pi }_{k}CUM\mathrm{\_}GROUP{S}_{s,t-k}+\gamma ln\left(INCIDENC{E}_{s,t}\right)+{\alpha }_s+{\delta }_t+{\epsilon }_{s,t}$(1) where the outcome measure is the number of years of life lost before age 85 (YPLL85) are shown in Figure 4 and Appendix Table 2. Each estimate is from a separate model. Solid squares indicate significant (p-value < .05) estimates; hollow squares indicate insignificant estimates. Dashed vertical lines represent 95% confidence intervals.

Figure 4. Estimates of the effect of the number of drug approvals in years 2000-k + 1 to 2017-k (k = 0, 2, 4,., 24) on the reduction inYPLL85 in 2017. Each estimate is from a separate model. Solid squares indicate significant (p-value < .05) estimates; hollow squares indicate insignificant estimates. Dashed vertical lines represent 95% confidence intervals

Table 2. Number of deaths and years of potential life lost before ages 85, 75, and 65 due to cancer, 2000–2017

Year	Deaths	YPLL85	YPLL75	YPLL65
2000	7,620	113,876	58,489	26,406
2005	7,864	113,662	59,099	26,762
2009	8,342	116,335	59,427	25,820
2013	8,936	120,519	60,659	25,391
2017	9,502	121,418	59,250	24,273

The estimates shown in Figure 4A and Panel A of Appendix Table 2 are from a model that includes CUM_DRUGS_s,t-k but does not include CUM_GROUPS_s,t-k. Figure 4A shows estimates of the percentage change in YPLL85 in 2017 due to new substances approved during the years 2000-k + 1 to 2017-k, calculated as exp(β_k * mean(ΔCUM_DRUGS_k)) – 1, where mean(ΔCUM_DRUGS_k) = mean(CUM_DRUGS_s,2017-k – CUM_DRUGS_s,2000-k), and mean() is the (weighted) mean across cancer sites. Only one of the estimates is significant when k ≤ 8, but all the estimates are significant when 10 ≤ k ≤ 24. This indicates that YPLL85 is at most weakly related to the number of substances approved less than 10 years earlier, but significantly inversely related to the number of substances approved at least 10 years earlier. This is not surprising, since as shown in Figure 3 utilization of a drug tends to be very low in the first 10 years after it was approved. YPLL85_s,t is most strongly inversely related to CUM_DRUGS_s,t-14. The estimates indicate that the substances approved during 1987–2003 reduced YPLL85 in 2017 by 43%.

The estimates shown in Figure 4B and Panel B of Appendix Table 2 are from a model that includes CUM_GROUPS_s,t-k but does not include CUM_DRUGS_s,t-k. In this case, only one of the estimates is significant when k ≤ 12, but all the estimates are significant when 14 ≤ k ≤ 24. Estimates of the effect on YPLL85 of the number of subgroups are larger than estimates of the effect of the number of substances. YPLL85_s,t is most strongly inversely related to CUM_GROUPS_s,t-20. The estimates indicate that the subgroups approved during 1981–1997 reduced YPLL85 in 2017 by 64%.

The estimates shown in Figure 4C and Panel C of Appendix Table 2 are from a model that includes both CUM_DRUGS_s,t-k and CUM_GROUPS_s,t-k. As discussed above, this model allows the effect on YPLL85 of approval of the first chemical substance in a chemical subgroup to differ from the effect of approval of a subsequent chemical substance in the chemical subgroup. In this case, none of the estimates is significant when k ≤ 10, but all the estimates are significant when 12 ≤ k ≤ 24. The magnitudes of the estimates in Figure 4C are similar to the magnitudes in Figure 4B. Once again, YPLL85_s,t is most strongly inversely related to the number of substances and subgroups approved 20 years earlier. The estimates indicate that the substances/subgroups approved during 1985–2001 reduced YPLL85 in 2017 by 67%.

Figure 4D juxtaposes the drug utilization profile from Figure 3 with the point estimates of the joint effect of substances and subgroups from Figure 4C. The shapes of the two curves are reasonably similar. The YPLL85-effect profile seems to lead the utilization profile, which may be due to the superior quality/efficacy of later-vintage drugs.

Estimates of eq. (1)(1) $ln\left(OUTCOM{E}_{s,t}\right)={\beta }_{k}CUM\mathrm{\_}DRUG{S}_{s,t-k}+{\pi }_{k}CUM\mathrm{\_}GROUP{S}_{s,t-k}+\gamma ln\left(INCIDENC{E}_{s,t}\right)+{\alpha }_s+{\delta }_t+{\epsilon }_{s,t}$(1) where the outcome measure is the number of years of life lost before age 75 (YPLL75) are shown in Appendix Figure 1 and Appendix Table 3. The overall pattern of the YPLL75 estimates is similar to the overall pattern of the YPLL85 estimates. Estimates of eq. (1)(1) $ln\left(OUTCOM{E}_{s,t}\right)={\beta }_{k}CUM\mathrm{\_}DRUG{S}_{s,t-k}+{\pi }_{k}CUM\mathrm{\_}GROUP{S}_{s,t-k}+\gamma ln\left(INCIDENC{E}_{s,t}\right)+{\alpha }_s+{\delta }_t+{\epsilon }_{s,t}$(1) where the outcome measure is the number of years of life lost before age 65 (YPLL65) are shown in Appendix Figure 2 and Appendix Table 4. The overall pattern of the YPLL65 estimates is also quite similar to the overall pattern of the YPLL85 estimates.

Table 3. Estimates of the effects of drug approvals on U.S. premature (before age 75) cancer mortality

Model	Parameter	Estimate	Standard Error	Z	Pr > |Z|
1	cum_drug (all drugs)	−0.0239	0.0061	−3.93	<.0001
2	cum_nz (drugs launched in both the U.S. and NZ)	−0.0246	0.0122	−2.03	0.0428
3	cum_not_nz (drugs launched in the U.S. and not launched in NZ)	−0.0236	0.0068	−3.48	0.0005

Estimates of eq. (1)(1) $ln\left(OUTCOM{E}_{s,t}\right)={\beta }_{k}CUM\mathrm{\_}DRUG{S}_{s,t-k}+{\pi }_{k}CUM\mathrm{\_}GROUP{S}_{s,t-k}+\gamma ln\left(INCIDENC{E}_{s,t}\right)+{\alpha }_s+{\delta }_t+{\epsilon }_{s,t}$(1) where the outcome measure is the odds of surviving at least 5 years after diagnosis (SURV5%_s,t/(1 – SURV5%_s,t)) are shown in Figure 5 and Appendix Table 5.¹⁴ The estimates shown in Figure 5A are estimates of (β_k * mean(ΔCUM_DRUGS_k)), not estimates of exp(β_k * mean(ΔCUM_DRUGS_k)) – 1; this also applies to Figure 5B and 5C. Figure 5A shows that the odds of surviving at least 5 years after diagnosis are significantly positively related to the number of substances approved 8 to 24 years before diagnosis; they are most strongly positively related to the number of substances approved 18 years earlier. In contrast to the YPLL results, the magnitude of the effect of the number of substances is larger than the magnitude of the effect of the number of subgroups. As noted earlier, between 1998 and 2010, the 5-year survival rate for all adult cancers increased from 57.7% to 63.3%. The estimates indicate that, if the number of drugs ever approved had not increased during 1981–1992, there would have been a statistically significant decline in the 5-year survival rate for all adult cancers. This could be attributable to an increase in the mean age of cancer patients. As shown in the Ministry of Health (2015a), older cancer patients have much lower relative survival rates.

Figure 5. Estimates of the effect of the number of drug approvals in years 1998‐k + 1 to 2010‐k (k = 0, 2, 4,., 24) on the odds of surviving at least 5 years after diagnosis in 2010. Each estimate is from a separate model. Solid squares indicate significant (p-value < .05) estimates; hollow squares indicate insignificant estimates. Dashed vertical lines represent 95% confidence intervals

Estimates of eq. (1)(1) $ln\left(OUTCOM{E}_{s,t}\right)={\beta }_{k}CUM\mathrm{\_}DRUG{S}_{s,t-k}+{\pi }_{k}CUM\mathrm{\_}GROUP{S}_{s,t-k}+\gamma ln\left(INCIDENC{E}_{s,t}\right)+{\alpha }_s+{\delta }_t+{\epsilon }_{s,t}$(1) where the outcome measure is the number of inpatient hospital discharges are shown in Figure 6 and Appendix Table 6. The estimates suggest that the number of hospital discharges may have been increased by new drugs approved at least 16 years earlier, possibly due to reduced mortality of cancer patients, but the evidence for this is quite weak: only one of the 13 estimates in Figure 6A, 6B, and 6C are statistically significant.

Figure 6. Estimates of the effect of the number of drug approvals in years 2005‐k + 1 to 2017‐k (k = 0, 2, 4,., 24) on the number of inpatient hospital discharges in 2017. Each estimate is from a separate model. Solid squares indicate significant (p-value < .05) estimates; hollow squares indicate insignificant estimates. Dashed vertical lines represent 95% confidence intervals

5. Discussion

As shown in Figure 4C, the drugs approved during 1985–2001 are estimated to have reduced YPLL85 in 2017 by 67%. This implies that if no drugs had been approved during 1985–2001, YPLL85 would have been 3.02 (= 1/(1–67%)) times as high as it actually was in 2017. YPLL85 from all cancers combined was 121,418 in 2017. Hence, if no drugs had been approved during 1985–2001, YPLL85 would have been 366,294 (= 3.02 * 121,418) in 2017; the drugs approved during 1985–2001 are estimated to have reduced YPLL85 in 2017 by 244,876 (= 366,294–121,418).¹⁵

In 2011, the Ministry of Health [31] estimated that the annual public price of all cancers was 511 million NZD. As shown in Appendix Table 7, that figure included the cost of public hospital discharge, outpatient attendance, community and hospital pharmacy, and other costs for all cancer patients from 1 year before registration to 5 years following registration. The Ministry also estimated that the price of cancer would increase by 23%, to 628 million NZD, by 2021. Even if previous drug approvals increased the cost of hospital discharges and other medical costs, the cost per life-year before age 85 gained in 2017 from those approvals could not have exceeded 2566 NZD (= 628 million NZD/244,876 life-years) or 1719 USD (at 0.67 USD/NZD). The World Health Organization considers interventions whose cost per quality-adjusted life-year (QALY) gained is less than 3 times per capita GDP to be cost-effective, and those whose cost per QALY gained is less than per capita GDP to be highly cost-effective [32,33]; New Zealand’s per capita GDP in 2017 was 42,260 USD.¹⁶

Data from the IQVIA New Product Focus database indicate that during the period 1986–2015, the number of cancer drugs launched in New Zealand was only half the number launched in the U.S. (68 vs. 139). In principle, it is possible that the drugs that were launched in the U.S. but not launched in New Zealand provide little or no benefit to patients, and therefore that New Zealand patients were not harmed by more limited access to new cancer drugs. We tested this hypothesis by estimating a model similar to eq. (1)(1) $ln\left(OUTCOM{E}_{s,t}\right)={\beta }_{k}CUM\mathrm{\_}DRUG{S}_{s,t-k}+{\pi }_{k}CUM\mathrm{\_}GROUP{S}_{s,t-k}+\gamma ln\left(INCIDENC{E}_{s,t}\right)+{\alpha }_s+{\delta }_t+{\epsilon }_{s,t}$(1) using U.S. data on outcomes and drug approvals, and by distinguishing between the effects on U.S. premature cancer mortality of (1) drugs launched in both the U.S. and New Zealand, and (2) drugs launched in the U.S. and not launched in New Zealand. We found that the two sets of drugs had almost identical (negative) effects on U.S. cancer mortality.¹⁷ This indicates that, in general, the drugs that were not launched in New Zealand were no less valuable than the drugs that were launched in New Zealand.

6. Summary and conclusions

We have analyzed the effect that pharmaceutical innovation – the introduction and use of new drugs used to treat cancer – had on the longevity and hospitalization of New Zealand cancer patients during the period 1998–2017, by investigating whether the cancer sites that experienced more pharmaceutical innovation had larger subsequent declines in premature (before ages 85, 75, and 65) mortality and hospitalization rates and larger subsequent increases in 5-year relative survival rates, controlling for changes in incidence. Due to imperfect data, our analysis is subject to several limitations. (1) We were unable to control for non-pharmaceutical medical innovation. (2) We do not have New Zealand-specific data on a drug’s initial or supplemental indications, or the timing of the latter. Therefore, we assumed that if a drug was indicated for a cancer site in October 2020 according to the Therapeutic Target Database, it was indicated for (and used to treat) cancer at that site in New Zealand when it was first approved in New Zealand. Pharmac may not subsidize a drug for all of the indications listed in the Therapeutic Target Database. (3) Our list of drugs used to treat different cancers may be incomplete. For example, according to the Therapeutic Target Database, zoledronic acid (M05BA08) is not used to treat cancer; its only indication is hypercalcemia (E83.5). However, I understand that zoledronic acid is used to prevent skeletal fractures in patients with cancers such as multiple myeloma and prostate cancer. (4) Although our econometric model allowed the effect on outcomes of the approval of the first chemical substance in a chemical subgroup to differ from the effect of a subsequent chemical substance in a chemical subgroup, we did not disaggregate substances in other dimensions, e.g. we did not distinguish between therapeutic and supportive agents, or between biologic/targeted agents and chemotherapy agents. This may be a fruitful area of future research. In the meantime, our estimates may capture the average effect of different types of agents.

YPLL85 is significantly inversely related to the number of chemical substances and subgroups approved at least 12 years earlier. (Utilization of a drug tends to be very low in the first 10 years after it was approved.) The estimates indicate that the substances/subgroups approved during 1985–2001 reduced YPLL85 in 2017 by 67%. Those substances/subgroups reduced YPLL75 and YPLL65 in 2017 by similar percentages.

The odds of surviving at least 5 years after diagnosis are significantly positively related to the number of substances approved 8 to 24 years before diagnosis; they are most strongly positively related to the number of substances approved 18 years earlier. Between 1998 and 2010, the 5-year survival rate for all adult cancers increased from 57.7% to 63.3%. The estimates indicate that, if the number of drugs ever approved had not increased during 1981–1992, there might have been a statistically significant decline in the 5-year survival rate for all adult cancers.

The estimates suggest that the number of hospital discharges may have been increased by new drugs approved at least 16 years earlier, possibly due to reduced mortality of cancer patients, but the evidence for this is quite weak.

The drugs approved during 1985–2001 are estimated to have reduced YPLL85 in 2017 by 244,876. Even if previous drug approvals increased the cost of hospital discharges and other medical costs, the cost per life-year before age 85 gained in 2017 from those approvals could not have exceeded 1719 USD. The World Health Organization considers interventions whose cost per quality-adjusted life-year gained is less than per capita GDP to be highly cost-effective; New Zealand’s per capita GDP in 2017 was 42,260 USD.

Declaration of interest

The author has received research support from Asociación de Laboratorios Farmacéuticos de Investigación y Desarrollo, Cámara Argentina de Especialidades Medicinales, Incyte Corporation, Korean Research-based Pharmaceutical Industry Association, Laerdal, Medicines Australia, Merck/MSD, National Pharmaceutical Council, Novartis, Pfizer, Pharmaceutical Research and Manufacturers of America, PHRMAG (Pharmaceutical Research and Manufacturers Association Gulf), PReMA (Thailand Pharmaceutical Research and Manufacturers Association), and the U.S. Social Security Administration. The author has no other relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript apart from those disclosed.

Reviewers disclosure

A reviewer on this manuscript has disclosed being the Deputy Editor in Chief, Journal of Medical Economics; Quantitative Methods Editor, JAMA Dermatology; Equity in Matrix45, LLC, which provides consultation services to the life sciences industries, government agencies (US, EU, international), academic institutions, and professional and patient advocacy organizations. Peer reviewers on this manuscript have no other relevant financial relationships or otherwise to disclose.

Additional information

Funding

Medicines New Zealand Inc. provided the author(s) with a research grant to perform the study. The author bears complete responsibility for the design and execution of the study.

Notes

1. Survival time for cancer patients is usually measured from the day the cancer is diagnosed until the day they die. Patients are often diagnosed after they have signs and symptoms of cancer. If a screening test leads to a diagnosis before a patient has any symptoms, the patient’s survival time is increased because the date of diagnosis is earlier. This increase in survival time makes it seem as though screened patients are living longer when that may not be happening. This is called lead-time bias. It could be that the only reason the survival time appears to be longer is that the date of diagnosis is earlier for the screened patients. But the screened patients may die at the same time they would have without the screening test [34].

2. SURV5%_s,t = the fraction of people diagnosed with cancer at site s in year t who were alive 5 years after diagnosis. The survival estimates are relative survival ratios. Relative survival is the observed survival in the cancer patient group divided by the expected survival of a matched group in the general population. Relative survival is cancer survival in the absence of other causes of death. It represents the proportion of patients within a particular group alive after a certain number of years of follow-up, most commonly five years, and attributes all the ‘excess’ mortality of the group to the cancer in question. For example, a relative survival of 75% means that the cancer reduces the likelihood of surviving five years after diagnosis by 25%. A relative survival of 100% indicates that cancer patients experience mortality rates equivalent to those in a comparable group from the general population. If a relative survival rate exceeds 100%, this indicates that cancer patients have better observed survival than is expected for people in the general population.

3. The vast majority (94%) of hospital discharges are publicly funded. In 2012–2013, there were 1.1 million publicly-funded discharges, and only 70 thousand privately-funded discharges.

4. In contrast, the age-adjusted mortality rate gives equal weight to reductions in mortality at different ages, e.g. reductions in the mortality rates of 85-year-olds and 40-year olds.

5. For example, the five levels associated with the chemical subgroup ‘nitrogen mustard analogues’ are:L ANTINEOPLASTIC AND IMMUNOMODULATING AGENTSL01 ANTINEOPLASTIC AGENTSL01AALKYLATING AGENTSL01AANitrogen mustard analoguesL01AA01cyclophosphamideL01AA02 chlorambucilL01AA03melphalanL01AA05chlormethineL01AA06ifosfamideL01AA07trofosfamideL01AA08prednimustineL01AA09bendamustine.

6. Some trends may have increased premature mortality. Between 1997 and 2014, the fraction of the New Zealand population that was obese increased from 18.8% to 29.9% [2].

7. N_CASES_s,t = the number of patients diagnosed with cancer at site s in year t.

8. mean(ΔCUM_DRUGS_k) is equal to the estimate of ρ₂₀₁₇ from the equation CUM_DRUGS_s,t-k = ϕ_s + ρ_t + ε_st. where ρ₂₀₀₀ is normalized to zero.

9. According to the Merriam-Webster dictionary, one definition of vintage is ‘a period of origin or manufacture (e.g. a piano of 1845 vintage)’. Robert Solow [35] introduced the concept of vintage into economic analysis. Solow’s basic idea was that technical progress is ‘built into’ machines and other goods and that this must be taken into account when making empirical measurements of their roles in production. This was one of the contributions to the theory of economic growth that the Royal Swedish Academy of Sciences cited when it awarded Solow the 1987 Alfred Nobel Memorial Prize in Economic Sciences [36].

10. Outpatient prescription drug claims usually don’t show the indication of the drug prescribed. Claims for drugs administered by doctors and nurses (e.g. chemotherapy) often show the indication of the drug, but these account for just 15% of drug expenditure. These data are not available for New Zealand.

11. A separate model is estimated for each value of k, rather than including multiple values (CUM_DRUGS_s,t, CUM_DRUGS_s,t-2, CUM_DRUGS_s,t-4, …) in a single model because CUM_DRUGS is highly serially correlated (by construction), which would result in extremely high multicollinearity if multiple values were included.)

12. The number of standard ‘dose’ units sold is determined by taking the number of counting units sold divided by the standard unit factor which is the smallest common dose of a product form as defined by IQVIA. For example, for oral solid forms, the standard unit factor is one tablet or capsule whereas for syrup forms the standard unit factor is one teaspoon (5 ml) and injectable forms it is one ampoule or vial. Other measures of quantity, such as the number of patients using the drug, prescriptions for the drug, or defined daily doses of the drug, are not available.

13. Mortality data are reported in 5-year age groups. We assume that deaths in a 5-year age group occur at the midpoint of the age group. For example, we assume that deaths at age 35–39 years occurred at age 37.5. The Association of Public Health Epidemiologists in Ontario [37] uses this method. These approximations result in some imprecision in the mortality estimates, but should not cause any bias in the parameter estimates.

14. The coefficient on the incidence measure (γ) was insignificant in all the 5-year survival odds models.

15. If the number of drugs approved during 1985–2001 had been 50% lower, rather than 100% lower, YPLL85 would have been 74% higher than it actually was in 2017; YPLL85 would have been 210,890. According to our model, pharmaceutical innovation is subject to diminishing marginal productivity: eliminating all innovation would have reduced YPLL85 more than twice as much as a 50% reduction in innovation.

16. Lichtenberg [38] demonstrated that the number of QALYs gained from pharmaceutical innovation could be either greater than or less than the number of life-years gained.

17. Table 3 shows estimates of the effects of drug approvals on U.S. premature (before age 75) cancer mortality.

Appendix

Figure 1. Estimates of the effect of the number of drug approvals in years 2000-k+1 to 2017-k (k = 0, 2, 4,., 24) on the reduction in YPLL75 in 2017

Figure 2. Estimates of the effect of the number of drug approvals in years 2000-k+1 to 2017-k (k = 0, 2, 4,., 24) on the reduction in YPLL65 in 2017

Table 1. MEDSAFE drug approvals for different cancer sites

ATC	Approval year	C00-C14	C15	C16	C18-C21	C22	C25	C33-C34	C43	C50	C51	C53	C54-C55	C56-C57	C61	C62	C64-C66, C68	C67	C71	C73	C81	C82-C85, C96	C90	C91-C95
Number approved		79	54	54	60	55	57	56	59	76	52	52	55	57	64	52	59	54	56	55	56	57	58	85
A01AC03 HYDROCORTISONE	1963	x
A03AE02 TEGASEROD	2002	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x
A07EA02 HYDROCORTISONE	1963	x
A07EA03 PREDNISONE	1966	x
A07EA04 BETAMETHASONE	1963	x
A14AB01 NANDROLONE	1969	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x
C03CA03 PIRETANIDE	1989																		x
C04AX02 PHENOXYBENZAMINE	1969														x
C05AA01 HYDROCORTISONE	1963	x
C05AA05 BETAMETHASONE	1963	x
C05AA11 FLUOCINONIDE	1971	x
D03AX05 HYALURONIC ACID	2000	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x
D05AD02 METHOXSALEN	1975																					x
D05BA02 METHOXSALEN	1975																					x
D06BB10 IMIQUIMOD	1998								x
D07AA02 HYDROCORTISONE	1963	x
D07AB10 ALCLOMETASONE	1984	x
D07AC01 BETAMETHASONE	1963	x
D07AC08 FLUOCINONIDE	1971	x
D07XA01 HYDROCORTISONE	1963	x
D07XC01 BETAMETHASONE	1963	x
D10AD01 TRETINOIN	1976																							x
G03AC01 NORETHISTERONE	1968	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x
G03AC05 MEGESTROL	1965									x			x
G03AC06 MEDROXYPROGESTERONE	1969	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x
G03BA01 FLUOXYMESTERONE	1969									x
G03BA02 METHYLTESTOSTERONE	1969	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x
G03CA03 ESTRADIOL	1969									x
G03CB02 DIETHYLSTILBESTROL	1969	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x
G03CC05 DIETHYLSTILBESTROL	1969	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x
G03DA02 MEDROXYPROGESTERONE	1969	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x
G03DA03 HYDROXYPROGESTERONE	1969	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x
G03DB02 MEGESTROL	1965									x			x
G03DB04 NOMEGESTROL	2012									x
G03DC02 NORETHISTERONE	1968	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x
G03EK01 METHYLTESTOSTERONE	1969	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x
G03HA01 CYPROTERONE	1976														x
G03XA01 DANAZOL	1979									x
G03XA02 GESTRINONE	1991									x
G03XB01 MIFEPRISTONE	2001																							x
G04BD08 SOLIFENACIN	2006																	x
G04BD10 DARIFENACIN	2006																	x
G04 CB02 DUTASTERIDE	2004														x
H01AB01 THYROTROPIN ALFA	1969																			x
H01CC02 CETRORELIX	2000													x
H02AB01 BETAMETHASONE	1963	x
H02AB07 PREDNISONE	1966	x
H02AB09 HYDROCORTISONE	1963	x
J07BH02 ROTA VIRUS, PENTAVALENT, LIVE, REASSORTED	2006	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x
L01AA01 CYCLOPHOSPHAMIDE	1969	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x
L01AA02 CHLORAMBUCIL	1969																							x
L01AA03 MELPHALAN	1969																						x
L01AA06 IFOSFAMIDE	1989	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x
L01AB02 TREOSULFAN	1984								x
L01AD02 LOMUSTINE	1980																		x
L01AD05 FOTEMUSTINE	1991	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x
L01AX03 TEMOZOLOMIDE	1998																		x
L01AX04 DACARBAZINE	1976								x
L01BA01 METHOTREXATE	1977	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x
L01BB02 MERCAPTOPURINE	1969																							x
L01BB03 TIOGUANINE	1974																							x
L01BB04 CLADRIBINE	1995																							x
L01BB05 FLUDARABINE	1993																				x	x		x
L01BB06 CLOFARABINE	2011																							x
L01BC01 CYTARABINE	1969																							x
L01BC02 FLUOROURACIL	1971	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x
L01BC05 GEMCITABINE	1998	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x
L01BC06 CAPECITABINE	2002				x					x
L01CA01 VINBLASTINE	1969	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x
L01CA02 VINCRISTINE	1969																							x
L01CA03 VINDESINE	1981																							x
L01CA04 VINORELBINE	1998	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x
L01CA05 VINFLUNINE	2011	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x
L01CB01 ETOPOSIDE	1982	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x
L01CB02 TENIPOSIDE	1982																							x
L01CD01 PACLITAXEL	1999	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x
L01CD02 DOCETAXEL	2008	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x
L01DA01 DACTINOMYCIN	1969	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x
L01DB01 DOXORUBICIN	1975	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x
L01DB02 DAUNORUBICIN	1971																							x
L01DB03 EPIRUBICIN	1984	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x
L01DB06 IDARUBICIN	1992																							x
L01DB07 MITOXANTRONE	1984	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x
L01DC01 BLEOMYCIN	1973																				x
L01DC02 PLICAMYCIN	1990	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x
L01DC03 MITOMYCIN	2010		x	x	x	x	x			x
L01DC04 IXABEPILONE	2008									x
L01XA01 CISPLATIN	1981	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x
L01XA02 CARBOPLATIN	1987													x
L01XA03 OXALIPLATIN	2004				x
L01XB01 PROCARBAZINE	1969																				x
L01XC02 RITUXIMAB	1999																				x	x
L01XC03 TRASTUZUMAB	2009									x
L01XC04 ALEMTUZUMAB	2009																							x
L01XC06 CETUXIMAB	2006				x
L01XC11 IPILIMUMAB	2012								x						x
L01XC13 PERTUZUMAB	2013									x
L01XC15 OBINUTUZUMAB	2014																							x
L01XE01 IMATINIB	2010	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x				x
L01XE02 GEFITINIB	2011	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x
L01XE03 ERLOTINIB	2011	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x
L01XE04 SUNITINIB	2006		x	x	x	x	x										x
L01XE05 SORAFENIB	2007	x				x	x	x	x					x			x							x
L01XE06 DASATINIB	2009																						x	x
L01XE07 LAPATINIB	2012									x
L01XE08 NILOTINIB	2010																							x
L01XE10 EVEROLIMUS	2005																x
L01XE11 PAZOPANIB	2010																x
L01XE13 AFATINIB	2014							x
L01XE15 VEMURAFENIB	2015								x											x
L01XE16 CRIZOTINIB	2015							x
L01XE17 AXITINIB	2013																x
L01XE18 RUXOLITINIB	2014						x
L01XE21 REGORAFENIB	2014				x
L01XE25 TRAMETINIB	2015								x
L01XX01 AMSACRINE	1984																							x
L01XX02 ASPARAGINASE	1971																							x
L01XX03 ALTRETAMINE	2000													x
L01XX05 HYDROXYCARBAMIDE	1969																							x
L01XX11 ESTRAMUSTINE	1979														x
L01XX14 TRETINOIN	1976																							x
L01XX17 TOPOTECAN	1998													x
L01XX19 IRINOTECAN	1997				x
L01XX24 PEGASPARGASE	2020	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x
L01XX27 ARSENIC TRIOXIDE	2019																							x
L01XX28 IMATINIB	2010																							x
L01XX32 BORTEZOMIB	2006																				x	x	x	x
L01XX45 CARFILZOMIB	2018																						x
L02AA01 DIETHYLSTILBESTROL	1969	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x
L02AA04 FOSFESTROL	1969	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x
L02AB01 MEGESTROL	1965									x			x
L02AB02 MEDROXYPROGESTERONE	1969	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x
L02AE01 BUSERELIN	1990	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x
L02AE02 LEUPRORELIN	1985														x
L02AE03 GOSERELIN	1987									x					x
L02AE04 TRIPTORELIN	1990	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x
L02BA01 TAMOXIFEN	1974									x
L02BA02 TOREMIFENE	1997									x
L02BA03 FULVESTRANT	2020									x
L02BB01 FLUTAMIDE	1986														x
L02BB02 NILUTAMIDE	1992														x
L02BB03 BICALUTAMIDE	2001														x
L02BG01 AMINOGLUTETHIMIDE	1981									x
L02BG02 FORMESTANE	1992									x
L02BG03 ANASTROZOLE	1996									x
L02BG04 LETROZOLE	1997									x
L02BG06 EXEMESTANE	2002									x
L03AA02 FILGRASTIM	1991	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x
L03AA13 PEGFILGRASTIM	2007	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x
L03AB03 INTERFERON GAMMA	1994	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x
L03AB11 PEGINTERFERON ALFA-2A	2003																							x
L03AC01 ALDESLEUKIN	2000																x					x
L03AX16 PLERIXAFOR	2015	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x
L04AA10 SIROLIMUS	2006																						x
L04AA18 EVEROLIMUS	2005																x
L04AX02 THALIDOMIDE	2003																						x
L04AX03 METHOTREXATE	1977	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x
L04AX04 LENALIDOMIDE	2008																						x
M05BX04 DENOSUMAB	2011														x
M09AX01 HYALURONIC ACID	2000	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x
N03AC02 TRIMETHADIONE	1965						x
N03AX12 GABAPENTIN	1994	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x
N05AB04 PROCHLORPERAZINE	1969							x							x				x
N06BA07 MODAFINIL	2004	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x
P02CE01 LEVAMISOLE	1993									x
R01AC08 OLOPATADINE	2002																							x
R01AD06 BETAMETHASONE	1963	x
R01AX09 HYALURONIC ACID	2000	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x
R03BA04 BETAMETHASONE	1963	x
R06AX17 KETOTIFEN	1984																							x
R06AX24 EPINASTINE	2004																							x
R07AX01 NITRIC OXIDE	1993	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x
S01BA02 HYDROCORTISONE	1963	x
S01BA06 BETAMETHASONE	1963	x
S01BA10 ALCLOMETASONE	1984	x
S01CB03 HYDROCORTISONE	1963	x
S01CB04 BETAMETHASONE	1963	x
S01GX04 NEDOCROMIL	1993																							x
S01GX08 KETOTIFEN	1984																							x
S01GX09 OLOPATADINE	2002																							x
S01GX10 EPINASTINE	2004																							x
S01KA01 HYALURONIC ACID	2000	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x
S01LA05 AFLIBERCEPT	2013				x
S02BA01 HYDROCORTISONE	1963	x
S02BA07 BETAMETHASONE	1963	x
S03AA07 CIPROFLOXACIN	1988																						x
S03BA03 BETAMETHASONE	1963	x
V04 CC04 CERULETIDE	1984	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x	x
V04 CJ01 THYROTROPIN	1969																			x
V10XX02 IBRITUMOMAB TIUXETAN (90Y)	2009																				x	x	x	x

Appendix 2

Table 2. Estimates of the effect of the no. of drug approvals in years 2000-k + 1 to 2017-k (k = 0, 2, 4,., 24) on YPLL85 in 2017

	ϕk = [βk * mean(ΔCUM_DRUGSk)]						$exp({\mathrm{\Phi }}_k)-1.$
k (lag)	LowerCL	Estimate	UpperCL	Std Err	Chi-Square	Pr > ChiSq	LowerCL	Estimate	UpperCL
0	−0.3555	−0.1672	0.0211	0.0961	3.03	0.0819	30%	15%	−2%
2	−0.3933	−0.1525	0.0882	0.1228	1.54	0.2142	33%	14%	−9%
4	−0.5542	−0.3144	−0.0746	0.1224	6.6	0.0102	43%	27%	7%
6	−0.5003	−0.2343	0.0317	0.1357	2.98	0.0842	39%	21%	−3%
8	−0.5423	−0.2449	0.0525	0.1517	2.61	0.1065	42%	22%	−5%
10	−0.6238	−0.3777	−0.1315	0.1256	9.04	0.0026	46%	31%	12%
12	−0.8375	−0.568	−0.2985	0.1375	17.06	<.0001	57%	43%	26%
14	−0.7595	−0.5683	−0.377	0.0976	33.92	<.0001	53%	43%	31%
16	−0.7827	−0.583	−0.3834	0.1019	32.76	<.0001	54%	44%	32%
18	−0.843	−0.5513	−0.2595	0.1488	13.72	0.0002	57%	42%	23%
20	−0.5806	−0.3687	−0.1569	0.1081	11.64	0.0006	44%	31%	15%
22	−0.7459	−0.4647	−0.1835	0.1435	10.49	0.0012	53%	37%	17%
24	−0.7218	−0.463	−0.2041	0.1321	12.29	0.0005	51%	37%	18%
	ϕk = [πk * mean(ΔCUM_GROUPSk)]						$exp({\mathrm{\Phi }}_k)-1.$
k (lag)	LowerCL	Estimate	UpperCL	Std Err	Chi-Square	Pr > ChiSq	LowerCL	Estimate	UpperCL
0	−0.312	−0.1683	−0.0247	0.0733	5.27	0.0217	27%	15%	2%
2	−0.4069	−0.1267	0.1534	0.1429	0.79	0.3752	33%	12%	−17%
4	−0.5702	−0.2559	0.0585	0.1604	2.54	0.1107	43%	23%	−6%
6	−0.4839	−0.1513	0.1813	0.1697	0.79	0.3726	38%	14%	−20%
8	−0.5493	−0.1605	0.2283	0.1984	0.65	0.4185	42%	15%	−26%
10	−0.7514	−0.2326	0.2862	0.2647	0.77	0.3796	53%	21%	−33%
12	−1.0137	−0.3807	0.2524	0.323	1.39	0.2386	64%	32%	−29%
14	−1.5083	−0.9247	−0.341	0.2978	9.64	0.0019	78%	60%	29%
16	−1.4991	−1.0236	−0.5481	0.2426	17.8	<.0001	78%	64%	42%
18	−1.4082	−0.983	−0.5577	0.217	20.52	<.0001	76%	63%	43%
20	−1.3398	−1.0043	−0.6688	0.1712	34.43	<.0001	74%	63%	49%
22	−1.0194	−0.7355	−0.4517	0.1448	25.79	<.0001	64%	52%	36%
24	−0.9574	−0.7004	−0.4433	0.1312	28.51	<.0001	62%	50%	36%
	ϕk = [βk * mean(ΔCUM_DRUGSk) + πk * mean(ΔCUM_GROUPSk)]						$exp({\mathrm{\Phi }}_k)-1$
k (lag)	LowerCL	Estimate	UpperCL	Std Err	Chi-Square	Pr > ChiSq	LowerCL	Estimate	UpperCL
0	−0.4011	−0.2151	−0.0291	0.0949	5.14	0.0234	33%	19%	3%
2	−0.4754	−0.1673	0.1407	0.1572	1.13	0.287	38%	15%	−15%
4	−0.6506	−0.3446	−0.0387	0.1561	4.87	0.0273	48%	29%	4%
6	−0.52	−0.2258	0.0684	0.1501	2.26	0.1324	41%	20%	−7%
8	−0.5447	−0.1343	0.2761	0.2094	0.41	0.5213	42%	13%	−32%
10	−0.7749	−0.2443	0.2863	0.2707	0.81	0.3668	54%	22%	−33%
12	−1.1733	−0.5932	−0.0132	0.296	4.02	0.045	69%	45%	1%
14	−1.4109	−0.9087	−0.4064	0.2562	12.57	0.0004	76%	60%	33%
16	−1.5441	−1.1042	−0.6644	0.2244	24.21	<.0001	79%	67%	49%
18	−1.5427	−1.0677	−0.5927	0.2423	19.41	<.0001	79%	66%	45%
20	−1.2951	−0.9737	−0.6524	0.164	35.27	<.0001	73%	62%	48%
22	−1.016	−0.7227	−0.4295	0.1496	23.33	<.0001	64%	51%	35%
24	−0.9409	−0.6897	−0.4386	0.1281	28.98	<.0001	61%	50%	36%

Appendix 3

Table 3. Estimates of the effect of the no. of drug approvals in years 2000-k + 1 to 2017-k (k = 0, 2, 4,., 24) on YPLL75 in 2017

	A. no. of substances only
	ϕk = [βk * mean(ΔCUM_DRUGSk)]						$exp({\mathrm{\Phi }}_k)-1.$
k (lag)	LowerCL	Estimate	UpperCL	Std Err	Chi-Square	Pr > ChiSq	LowerCL	Estimate	UpperCL
0	−0.3539	−0.1349	0.084	0.1117	1.46	0.2271	30%	13%	−9%
2	−0.3893	−0.1084	0.1726	0.1434	0.57	0.4497	32%	10%	−19%
4	−0.5827	−0.3025	−0.0223	0.143	4.48	0.0343	44%	26%	2%
6	−0.5451	−0.2232	0.0987	0.1643	1.85	0.1742	42%	20%	−10%
8	−0.5556	−0.241	0.0735	0.1605	2.26	0.1331	43%	21%	−8%
10	−0.6238	−0.359	−0.0941	0.1351	7.06	0.0079	46%	30%	9%
12	−0.8811	−0.6073	−0.3334	0.1397	18.89	<.0001	59%	46%	28%
14	−0.8284	−0.6054	−0.3823	0.1138	28.29	<.0001	56%	45%	32%
16	−0.8783	−0.6136	−0.3489	0.135	20.65	<.0001	58%	46%	29%
18	−0.8395	−0.5912	−0.3428	0.1267	21.77	<.0001	57%	45%	29%
20	−0.5677	−0.3728	−0.1779	0.0994	14.06	0.0002	43%	31%	16%
22	−0.8067	−0.5624	−0.3181	0.1246	20.36	<.0001	55%	43%	27%
24	−0.7936	−0.5849	−0.3761	0.1065	30.15	<.0001	55%	44%	31%
				B. no. of subgroups only
	ϕk = [πk * mean(ΔCUM_GROUPSk)]						$exp({\mathrm{\Phi }}_k)-1$
k (lag)	LowerCL	Estimate	UpperCL	Std Err	Chi-Square	Pr > ChiSq	LowerCL	Estimate	UpperCL
0	−0.3599	−0.1415	0.0769	0.1114	1.61	0.2041	30%	13%	−8%
2	−0.4132	−0.0443	0.3247	0.1882	0.06	0.8141	34%	4%	−38%
4	−0.6605	−0.3138	0.0329	0.1769	3.15	0.0761	48%	27%	−3%
6	−0.6055	−0.225	0.1554	0.1941	1.34	0.2463	45%	20%	−17%
8	−0.6284	−0.1799	0.2687	0.2289	0.62	0.4319	47%	16%	−31%
10	−0.7434	−0.168	0.4073	0.2935	0.33	0.567	52%	15%	−50%
12	−1.4269	−0.6339	0.1591	0.4046	2.45	0.1172	76%	47%	−17%
14	−1.8757	−1.223	−0.5703	0.333	13.49	0.0002	85%	71%	43%
16	−1.6243	−1.1225	−0.6207	0.256	19.22	<.0001	80%	67%	46%
18	−1.6318	−1.1214	−0.6109	0.2604	18.54	<.0001	80%	67%	46%
20	−1.3928	−1.0166	−0.6404	0.1919	28.06	<.0001	75%	64%	47%
22	−1.3622	−0.9792	−0.5961	0.1954	25.1	<.0001	74%	62%	45%
24	−1.1891	−0.8744	−0.5597	0.1606	29.66	<.0001	70%	58%	43%
					C. no. of substances & subgroups
	ϕk = [βk * mean(ΔCUM_DRUGSk) + πk * mean(ΔCUM_GROUPSk)]						$exp({\mathrm{\Phi }}_k)-1$
k (lag)	LowerCL	Estimate	UpperCL	Std Err	Chi-Square	Pr > ChiSq	LowerCL	Estimate	UpperCL
0	−0.4082	−0.1746	0.0591	0.1192	2.14	0.1431	34%	16%	−6%
2	−0.446	−0.0734	0.2993	0.1901	0.15	0.6995	36%	7%	−35%
4	−0.7002	−0.3804	−0.0607	0.1632	5.44	0.0197	50%	32%	6%
6	−0.6064	−0.2759	0.0545	0.1686	2.68	0.1017	45%	24%	−6%
8	−0.6252	−0.1496	0.326	0.2427	0.38	0.5375	46%	14%	−39%
10	−0.7881	−0.1883	0.4114	0.306	0.38	0.5383	55%	17%	−51%
12	−1.5807	−0.8583	−0.1358	0.3686	5.42	0.0199	79%	58%	13%
14	−1.8388	−1.1769	−0.5149	0.3377	12.14	0.0005	84%	69%	40%
16	−1.6415	−1.2016	−0.7616	0.2245	28.65	<.0001	81%	70%	53%
18	−1.7416	−1.2132	−0.6848	0.2696	20.25	<.0001	82%	70%	50%
20	−1.3558	−0.9799	−0.6041	0.1918	26.11	<.0001	74%	62%	45%
22	−1.382	−0.9744	−0.5669	0.2079	21.96	<.0001	75%	62%	43%
24	−1.1716	−0.8591	−0.5467	0.1594	29.04	<.0001	69%	58%	42%

Appendix 4

Table 4. Estimates of the effect of the no. of drug approvals in years 2000-k + 1 to 2017-k (k = 0, 2, 4,., 24) on YPLL65 in 2017

	A. no. of substances only
	${\mathrm{\Phi }}_k=[{\beta }_k\ast mean(\mathrm{\Delta }CUM\mathrm{\_}DRUG{S}_k)]$						$exp({\mathrm{\Phi }}_k)-1$
k (lag)	LowerCL	Estimate	UpperCL	Std Err	Chi-Square	Pr > ChiSq	LowerCL	Estimate	UpperCL
0	−0.4184	−0.0336	0.3512	0.1963	0.03	0.8642	34%	3%	−42%
2	−0.5324	0.0217	0.5757	0.2827	0.01	0.9389	41%	−2%	−78%
4	−0.7132	−0.2369	0.2393	0.243	0.95	0.3295	51%	21%	−27%
6	−0.6945	−0.1931	0.3083	0.2558	0.57	0.4503	50%	18%	−36%
8	−0.6285	−0.206	0.2165	0.2156	0.91	0.3393	47%	19%	−24%
10	−0.7034	−0.311	0.0813	0.2002	2.41	0.1202	51%	27%	−8%
12	−0.9829	−0.6653	−0.3477	0.162	16.86	<.0001	63%	49%	29%
14	−0.9501	−0.6408	−0.3315	0.1578	16.49	<.0001	61%	47%	28%
16	−1.0842	−0.6598	−0.2353	0.2166	9.28	0.0023	66%	48%	21%
18	−0.9769	−0.6486	−0.3203	0.1675	15	0.0001	62%	48%	27%
20	−0.6576	−0.3809	−0.1041	0.1412	7.27	0.007	48%	32%	10%
22	−1.0904	−0.7323	−0.3742	0.1827	16.06	<.0001	66%	52%	31%
24	−1.1107	−0.7785	−0.4464	0.1695	21.11	<.0001	67%	54%	36%
					B. no. of subgroups only
k (lag)	LowerCL	Estimate	UpperCL	Sd Err	Chi-Square	Pr > ChiSq	LowerCL	Estimate	UpperCL
0	−0.4504	0.0168	0.484	0.2384	0	0.9437	36%	−2%	−62%
2	−0.4329	0.2774	0.9877	0.3624	0.59	0.444	35%	−32%	−169%
4	−0.8735	−0.2089	0.4557	0.3391	0.38	0.5378	58%	19%	−58%
6	−0.7972	−0.2121	0.3731	0.2986	0.5	0.4775	55%	19%	−45%
8	−0.7437	−0.0985	0.5467	0.3292	0.09	0.7648	52%	9%	−73%
10	−0.9333	0.1072	1.1477	0.5309	0.04	0.84	61%	−11%	−215%
12	−1.9612	−0.9175	0.1261	0.5325	2.97	0.0849	86%	60%	−13%
14	−2.3316	−1.371	−0.4104	0.4901	7.83	0.0052	90%	75%	34%
16	−1.8521	−1.0141	−0.1761	0.4276	5.63	0.0177	84%	64%	16%
18	−1.6261	−0.9961	−0.3661	0.3214	9.6	0.0019	80%	63%	31%
20	−1.5315	−0.9164	−0.3012	0.3139	8.52	0.0035	78%	60%	26%
22	−2.0684	−1.432	−0.7956	0.3247	19.45	<.0001	87%	76%	55%
24	−1.7731	−1.1913	−0.6095	0.2969	16.1	<.0001	83%	70%	46%
					C. no. of substances & subgroups
	ϕk = [βk * mean(ΔCUM_DRUGSk) + πk * mean(ΔCUM_GROUPSk)]						$exp({\mathrm{\Phi }}_k)-1$
k (lag)	LowerCL	Estimate	UpperCL	Std Err	Chi-Square	Pr > ChiSq	LowerCL	Estimate	UpperCL
0	−0.5001	−0.0043	0.4916	0.253	0	0.9865	39%	0%	−63%
2	−0.4671	0.2581	0.9833	0.37	0.49	0.4854	37%	−29%	−167%
4	−0.9005	−0.2615	0.3774	0.326	0.64	0.4224	59%	23%	−46%
6	−0.8097	−0.2489	0.3119	0.2861	0.76	0.3843	56%	22%	−37%
8	−0.7072	−0.0583	0.5906	0.3311	0.03	0.8602	51%	6%	−81%
10	−0.9879	0.0772	1.1422	0.5434	0.02	0.8871	63%	−8%	−213%
12	−2.1209	−1.1657	−0.2104	0.4874	5.72	0.0168	88%	69%	19%
14	−2.3458	−1.3131	−0.2803	0.5269	6.21	0.0127	90%	73%	24%
16	−1.8117	−1.1118	−0.4119	0.3571	9.69	0.0018	84%	67%	34%
18	−1.7678	−1.13	−0.4923	0.3254	12.06	0.0005	83%	68%	39%
20	−1.5248	−0.8469	−0.1689	0.3459	5.99	0.0143	78%	57%	16%
22	−2.1057	−1.4385	−0.7712	0.3404	17.85	<.0001	88%	76%	54%
24	−1.7325	−1.1713	−0.61	0.2863	16.73	<.0001	82%	69%	46%