Impact of Oncology Drug Shortages on Chemotherapy Treatment

Abstract

Prescription drug shortages began to increase markedly in the mid-2000s, including sterile injectable products such as chemotherapy drugs. Using Medicare claims linked to Surveillance Epidemiology and End Results (SEER), we examined outpatient chemotherapy use during shortage periods relative to the months before and after a shortage for newly diagnosed patients with breast, colorectal, leukemia, lung, lymphoma, ovarian or pancreatic cancer (N=182,470). For most drugs, we found little impact of shortages on either the fraction of patients receiving that drug or the quantity provided. In some cases, we found declines in utilization: 4% for doxorubicin and fluorouracil; 2.9% for oxaliplatin; and about 1% for cytarabine, dacarbazine and leuprolide. While shortages for a few drugs resulted in substantial reductions in use, in most cases they resulted in little to no reduction. We discuss potential explanations for these counterintuitive findings, including potential limitations of current drug shortage reporting methods.

INTRODUCTION

The number of prescription drug shortages in the United States increased from 71 in 2005 to 255 in 2011, following many years of relatively few shortages^1,2 While the number of new shortages has fallen since 2012, the duration of existing shortages has increased.³ In the first quarter of 2015, 265 drugs remained in short supply.⁴ Many drugs in shortage are sterile injectable products, such as chemotherapy, anesthesia and anti-infective agents.^5,6 Shortages of oncology drugs more than doubled between 2005 and 2011 (see Figure 1).

Figure 1.

Frequency of Oncology Drug Shortages: 2001–2012 Data Source: UUDIS

Oncology drug shortages have raised specific alarm because of the potentially serious consequences for patient health and survival from inadequate drug dosing and treatment gaps.⁷ For cancers where certain regimens are clinically preferred or no good therapeutic alternative exists, drug shortages may be life threatening.^8,9 For example, cytarabine, a cornerstone of acute myeloid leukemia (AML) treatment that generates expected survival rates of over 50% and is without a substitute, was in shortage for much of 2011.¹⁰ Consequently, some leading cancer centers rationed supplies and some hospitals stopped offering the treatment.¹¹

In response to oncology and other critical drug shortages, policymakers have drafted reports to understand the cause of shortages,^1,2 issued an executive order to encourage early drug shortage warnings and expedited review of new manufacturing facilities, and expanded the authority of the Food and Drug Administration (FDA) under the FDA Safety and Innovation Act (FDASIA) of 2012 to respond to shortages.^3,12 Despite substantial policy and media attention, however, the consequences of drug shortages are still not well understood. To date, no studies have systematically analyzed the consequences of shortages nationwide. Most prior work is based on survey opinions^13,14 or case-studies of specific drugs, hospitals or health systems.^{15,16,17,18,19} Most of these studies find that shortages adversely affect treatment and health outcomes and increase spending. For example, a shortage of mechlorethamine led to a decline in 2-year event-free survival in children treated at Stanford University for Hodgkin’s lymphoma—from 88% to only 75% for those treated with the same regimen but with cyclophsphamide as a substitute for mechlorethamine.¹⁷

While prior studies contribute to our understanding of the local effects of shortages or the larger scale effects of specific drug shortages, they may not be representative.²⁰ The tendency to favor “positive” over null findings suggests these studies may provide a skewed view.²¹ Using data from the Surveillance, Epidemiology, and End Results (SEER) cancer registries linked with Medicare claims, we aimed to provide a more general analysis of the impact of oncology drug shortages on outpatient chemotherapy treatment.

RESULTS

Study Population

Our sample included 2,457,177 monthly claims for chemotherapy treatment administered in a physician’s office or outpatient hospital or clinic setting from 2004–2011 covering 182,470 Medicare beneficiaries newly diagnosed with cancer (Breast=51,166; Colon and Rectum=29,992; Leukemia=5,038; Lung=56,779; Lymphoma=21,596; Ovary=6,805; Pancreas=11,094).²² The newly diagnosed sample represented 60% of total chemotherapy claims in the SEER-Medicare data. 33.6% of chemotherapy claims for the newly diagnosed were for drugs that were on shortage during the study period.

Effect of Shortages on Receiving Treatment

Figure 2 Panel A shows estimated effects of shortages on the fraction of patients receiving treatment for each chemotherapy drug.²³ The fraction of patients treated declined by more than 4%²⁴ during shortage months for doxorubicin [95% CI, −7.0% to −1.1%] and fluorouracil [95% CI,−5.6% to −3.0%] and by 2.9% for oxaliplatin [95% CI, −5.3% to −0.42%]; respective p-values were 0.038, 0.019 and 0.064 after adjustment for multiple comparisons (see Table S1, Panel A). Smaller (1%) reductions in treatment were found for three other drugs: cytarabine [95% CI,−1.7% to −0.3%], dacarbazine [95% CI,−1.4% to −0.3.%] and leuprolide [95% CI,−1.5% to −0.6%], with p-values of 0.038, 0.032 and 0.0095 after adjustment for multiple comparisons.

Figure 2.

Percentage Change in Utilization During a Shortage

Panel A: Fraction of Patients Treated

Panel B: Units Per Patient

Notes: Shortages are defined using UUDIS coding. The points represent the coefficient estimate for the shortage effect from the interrupted time series regression using a linear model with IHS transformation (see shortage indicator in Equation 1 of the Supplementary Materials). The bars are 95% confidence intervals for each coefficient. Points highlighted in orange represent the most commonly used drugs (i.e., they are among the top 25 chemotherapy drugs in the Medicare claims). Four drugs (capecitabine, dactinomycin, denileukin, interferon alfa 2A) are not included in the figure because units are missing during shortage months. Missing units occur when the drug is given in an outpatient hospital setting rather than a physician’s office.

For the remaining 32 out of 38 drugs experiencing shortages, we found either no statistically significant effect of shortages on treatment (after adjusting for multiple comparisons) or, in a few cases, an increase in treatment (see Figure S1 for specific case studies). Even among the most commonly used chemotherapy drugs – those ranked in the top 25 prescribed drugs in our sample– we found almost no change in treatment during shortage episodes. The exception among commonly used drugs was etoposide, which increased by an estimated 6% [95% CI, 2.2% to 10.5%], with a p-value of 0.03 after adjusting for multiple comparisons.

Our estimates for commonly used drugs were largely insensitive to modeling choice. Results were similar when we estimated absolute changes in utilization using conventional linear models without the IHS transformation (see Figure S2). Similarly, using the Poisson regression model (see Table S2), we estimated a 5.1% [95% CI, −8.8% to −1.3%] reduction during shortage months for doxorubicin compared to a 4.1% [95% CI, −7.0% to −1.1%] reduction using the IHS transformation. The Poisson versus IHS estimates were −4.5% [95% CI, −5.8% to −3.1%] versus −4.3% [95% CI,−5.6% to −3.0%] for fluorouracil and −4.3% [95% CI, −7.5% to −1.0%] versus −2.9% [95% CI, −5.3% to −0.42%] for oxaliplatin treatment. The point estimates were more sensitive to modeling choice for infrequently used drugs.

The duration of shortages was not strongly predictive of the magnitude of the utilization effect (see Figure S3), as we observed the largest reductions in treatment for drugs that were on shortage for 5 (oxaliplatin), 26 (doxorubicin), and 73 (fluorouracil) months during our study period. Overall, we saw a similar proportion of treatment reductions across shortage durations, although the negative effects were somewhat more concentrated among drugs on shortage for more than 10 months.

Effects of Shortages on Quantity of Treatment

We found no effect of shortage episodes on dosing volume for the most commonly used drugs. This can be seen in Figure 2 Panel B, which shows percentage changes in the number of units billed per treated patient. Although the fraction of patients receiving treatment with some commonly used drugs, such as fluorouracil and doxorubicin, declined during shortage periods, the dosing of these drugs were not affected. In contrast, dosing for some less frequently used drugs without a single clinically equivalent therapeutic substitute – such as carmustine and cytarabine^25,26 – was reduced. For drugs with few substitutes, clinicians may have adjusted to shortages by limiting dosages rather than treatments.

Alternate Coding of Shortages

We tested the sensitivity of our results to an alternate coding of shortage episodes – those identified by the FDA. Since the FDA is focused on market-wide disruptions, it captured many fewer shortages than the UUDIS. Figure 3 Panel A shows the estimated reduction in utilization and dosing for drugs on the FDA shortage list. We found statistically significant reductions in use for only 5 of 19 drugs on the FDA shortage list: etoposide, −8.8% [95% CI, −13% to −4.8%]; fluorouracil, −6.7%, [95% CI, −9.2% to −4.1%], leucovorin calcium, −8.3% [95% CI, −12.5% to −4.1%], methotrexate, −1.0% [95% CI, −1.8% to −0.26%], and vinblastine −2.1% [95% CI, −3.75% to −0.52%]. Adjusting for multiple comparisons, the p-values were 0.006, 0.0095, 0.005, 0.029 and 0.038, respectively. For etoposide, this reduction contrasts with an increase in use found with the UUDIS shortage dates. The sign of the estimated change in use flips for several other drugs, including doxorubicin. These differences reflect discrepancies in the dating of shortages across data sources. For example, the FDA dated the start of the doxorubicin shortage 6 months after UUDIS. The pattern of results for dosing was similar (see Figure 3 Panel B), with dosages declining for a few drugs, remaining unchanged for most and even increasing for others.

Figure 3.

Percentage Changes in Utilization During a Shortage Using FDA Shortage Definitions

Panel A: Fraction of Patients Treated

Panel B: Units Per Patient

Notes: Shortages are defined using FDA coding. The points represent the coefficient estimate for the shortage effect from the interrupted time series regression using a linear model with IHS transformation (see shortage indicator in Equation 1 of the Supplementary Materials). The bars are 95% confidence intervals for each coefficient. Points highlighted in orange represent the most commonly used drugs (i.e., they are among the top 25 chemotherapy drugs in the Medicare claims). Three drugs (asparaginase, capecitabine and daunorubicin) are not included in Panel B because units are missing during FDA shortage months. Missing units occur when a drug is in an outpatient hospital setting rather than a physician’s office.

DISCUSSION

We found little impact on outpatient chemotherapy treatment for the majority of oncology drugs identified as experiencing shortages between 2004–2011. While the proportion of patients receiving treatment declined for six drugs, including fluorouracil, doxorubicin, and cytarabine, which have been prominently featured in the media,^27,28,29,30 the use of most other drugs was unaffected or even increased during shortage episodes. Likewise, dosages declined for only a few drugs during shortages. These findings are surprising in light of the substantial media and policy attention that the drug shortage problem has garnered. There are a few possible explanations for our counterintuitive findings.

First, our findings may indicate that most oncology drug shortages have been effectively managed to minimize the effects on treatment, even if at nontrivial personnel and psychic costs.³¹ While a shortage may reflect a complete disruption of production, providers may be able to draw from existing inventories, purchase from (or direct patients to) providers who have higher projected inventory or stockpile in advance if the shortage is foreseen. Providers may have successfully implemented shortage management strategies, several of which have been discussed in practitioner journals.^32,33

Second, rapid or timely production increases, sometimes at the request of the FDA, could have mitigated the impact of shortages on chemotherapy treatment. In some cases, temporary FDA allowance of importation of unapproved foreign versions of scarce drugs may have further eased the effects of shortages.³⁴

Third, the UUDIS database (the gold standard for dating shortages) may be capturing many shortages that would not be expected to impact treatment. Consistent with its primary objective of supporting physicians and pharmacies in adjusting their practices, the UUDIS includes on its shortage list any supply disruption that affects “how a pharmacy prepares or dispenses a drug product or that influences patient care when prescribers must use an alternative agent.”¹⁹ For example, if a particular presentation of a drug (e.g., larger vial size) were unavailable, this would be classified as a shortage even if other package sizes (e.g., smaller vial size) were still being supplied and met patient demand (e.g., see the case of Vancomycin¹⁹). While this type of supply disruption might require additional pharmacy resources, its impact on treatment might be minimal. It is unclear how many of the UUDIS shortages are of this variety, however our evidence of largely null shortage impacts suggest that many may fit this definition.

In contrast, the FDA considers a drug on shortage only if the combined supply from all manufacturers in a specific drug market cannot meet historical demand.¹⁹ As a result, the FDA has many fewer drugs on its shortage list than the UUDIS.³⁵ The more conservative coding may account for the higher proportion of shortages that were estimated to affect utilization. On the other hand, the FDA reports different shortage durations (see Table S3), generally dating them later and ending them earlier than the UUDIS since UUDIS does not consider a shortage resolved until all forms of the drug are back on the market. The difference in the timing of shortages may explain the markedly different estimates for a few drugs. While the UUDIS and FDA provide the best available data on shortage episodes, mis-measured shortage episodes could lead us to under or overstate the impact of a shortage.

Finally, our findings may be due to SEER-Medicare data limitations. While the key advantages of these data for our study are that it provides a large sample size, relevant population of cancer patients, and high frequency data to measure outcomes before and after shortage episodes, it captures relatively few non-elderly individuals.³⁶ Thus, it may understate the effect of shortages on drugs, such as mechlorethamine, used to treat cancers that primarily affect younger populations. Also, while SEER-Medicare is geographically diverse and captures fee-for-service beneficiaries in areas representing about 30% of the US population,³⁷ the data may not accurately capture the experience of shortages nationwide, which is a particular concern if there is variation in the shortage effect across areas.³⁸ That is, the data cannot capture treatment changes outside of the SEER registry areas nor among the increasing share of beneficiaries in Medicare Advantage plans. Finally, because some of the drugs in our sample are used infrequently, we may not have the power to fully measure the consequences of shortages for these drugs. Nonetheless, even among commonly used drugs, we found virtually no measurable impact of shortages on treatment.

The number of reported shortages by the UUDIS and the FDA increased dramatically over the last decade. For providers, shortages undoubtedly disrupt practice. For patients needing chemotherapy treatment, a shortage adds anxiety to an already stressful and complicated experience. It may even impact the patient’s course of treatment and, in some instances, survival. On average, however, we detected significant changes to outpatient chemotherapy treatment in only a few select cases. In a few such cases, however, the impacted drug is a cornerstone of therapy for specific cancers (e.g., doxorubicin for non-Hodgkin’s lymphoma and certain breast and ovarian cancers or cytarabine for acute myeloid leukemia), implying potentially important consequences from sustained reductions in use. In other cases, close substitutes exist (e.g., capecitabine for fluorouracil) such that the consequences of reductions in use may be more muted.

Most shortages, whether identified by the UUDIS or the FDA, seemed to have limited effects on outpatient utilization or dosing. An important caveat is that Medicare, and to our knowledge most commercial, claims data do not identify specific inpatient chemotherapy agents. Thus, while our analysis is informative about the majority of adult cancers treated in the outpatient setting, it cannot shed light on cancers typically treated in the inpatient setting, most notably leukemias.

A key challenge moving forward is how to separately identify those cases that will result in clinically relevant shortages from those that are relatively easy to manage. More direct articulation of the sources of, intended use for, and limitations of the UUDIS and FDA drug shortage lists would improve decision-making by policymakers and physicians and the public discussion of drug shortages. Finally, alternative approaches to tracking shortages and measuring their effects on treatment may be warranted to facilitate the early identification and mitigation of clinically relevant shortages.

METHODS

Data

We measured drug shortage episodes using the University of Utah Drug Information Service (UUDIS) database of shortages reported to the American Society of Health-System Pharmacists (ASHP) from 2001 through 2012. These data have been used in several key government reports documenting the rise in the frequency of drug shortages.^1,2,5 They are widely considered to be the most comprehensive and reliable source of information on drug shortages, since UUDIS has been tracking shortages in a consistent way for over a decade.³⁹

From the UUDIS, we selected drugs with an American Hospital Formulary System (AHFS) number identifying antineoplastic agents. We identified 90 shortage episodes involving 39 distinct antineoplastic agents between 2004 and 2011, our main study period. In sensitivity analyses, we use the FDA’s drug shortage list, which captures a more limited set of drug shortages that the agency characterizes as market-wide supply disruptions. We included drugs on the FDA shortage list that were identified as antineoplastic agents on the UUDIS list (see the Supplementary Materials for more details on the shortage data).

We linked oncology drug shortage episodes to cancer treatment data from Medicare-linked SEER cancer registries, considered a gold standard for cancer health services research (Dusetzina et al. 2014). The data cover all Medicare beneficiaries with cancer residing in SEER geographic areas (CA, CT, HI, IA, KY, LA, NJ, NM, UT and parts of GA, MI, and WA). The SEER-Medicare data provides detailed information on outpatient chemotherapy treatments administered in the physician’s office, outpatient hospital setting and at home through Medicare’s coverage of durable medical equipment. The claims include the Healthcare Common Procedure Coding System (HCPCS) number, which allowed us to identify the drug names used in each treatment. For the majority of claims, we also observe the total billing units of each drug administered; however, this information is not available for outpatient hospital claims due to bundled institutional billing under the outpatient prospective payment system (OPPS).

We studied the SEER-Medicare cohort that was diagnosed with breast, colorectal, leukemia, lung, lymphoma, ovarian and pancreatic cancers between 2003 and 2011 and all associated Medicare drug claims for these individuals from 2002 to 2012. These were the most currently available data from SEER-Medicare at the time of this study and captured the peak of oncology shortages in 2010 and 2011. These cancers account for over 45 percent of all new cancer cases nationally⁴⁰ and are commonly treated with chemotherapy, including many of the drugs subject to shortages over this period.

Sample Restrictions

To capture treatment, we restricted our claims analysis to 2004 to 2011, which enabled us to create a balanced sample of individuals newly diagnosed with cancer. We defined newly diagnosed patients in each month as those who had been diagnosed with cancer 0–11 months prior to the observation month (e.g., to measure treatment outcomes in January 2004, we include individuals diagnosed with cancer between February 2003-January 2004). We excluded beneficiaries with more than one primary cancer and those enrolled in a health maintenance organization (HMO) at any point during the study period. We further required that beneficiaries be enrolled in Part B, since the vast majority of chemotherapy is provided in outpatient settings. We also excluded patients whose dates of diagnosis or death differed by more than two months in the SEER and Medicare claims databases, whose month of diagnosis is unknown, or whose cancer was first identified at the time of death.

We aggregated outpatient chemotherapy claims by month and drug to generate a sample of 8,656 drug-by-month observations across all antineoplastic agents. Our final sample includes 3,577 drug-by-month observations of agents ever in shortage during the study period.

Statistical Analysis

We used an interrupted time series regression model to compare mean utilization of each chemotherapy drug during shortage periods relative to the months before and after a shortage. Our primary outcome was the fraction of patients receiving treatment with each drug among those who were eligible for treatment. Eligibility is based on having been diagnosed with a cancer that has ever been treated with that drug. A secondary outcome was the mean number of units billed for each drug for patients receiving chemotherapy. We estimated the effect of shortages as the average deviation in drug utilization during shortage months relative to the average drug-specific utilization trend across all months. We estimated this shortage impact separately for each drug.

We estimated all models using linear regression but transformed our outcome variables using the inverse hyperbolic sine (IHS) to account for zeros in the utilization data.^41,42 In sensitivity analyses, we estimated both linear models without the IHS transformation and Poisson regression models. The latter model also accounts for zeros in the dependent variable.⁴³ For the IHS and Poisson models, we reported the percentage change in the outcome during shortage relative to non-shortage periods.

We reported confidence intervals using two-tailed tests and a 0.05 significance level. The confidence intervals are based on heteroskedastic robust standard errors. To account for multiple comparisons across the many drugs experiencing shortages, we also reported p-values based on Benjamini–Hochberg adjustments to account for the false discovery rate in Table S1 (see the Suplementary Materials for additional details on the analysis, including our estimating equations).⁴⁴

STUDY HIGHLIGHTS.

What is the current knowledge on the topic?

To date, no studies have systematically analyzed the consequences of oncology drug shortages nationwide.

What question did this study address?

This study asked how oncology drug shortages have impacted chemotherapy use in Medicare beneficiaries newly diagnosed with seven relatively common cancers (breast, colorectal, leukemia, lung, lymphoma, ovarian or pancreatic cancer).

What does this study add to our knowledge?

Our study finds that in most cases, reported oncology drug shortages have a relatively limited effect on utilization. In a few cases, these shortages had quite substantial impacts on utilization. One cause of this surprising finding may be that most shortages reported, while critical for pharmacy workflow and productivity, may have less noticeable impacts from the patient perspective.

How might this change clinical pharmacology or translational science?

More direct articulation of the sources of, intended use for, and limitations of the UUDIS and FDA drug shortage lists could potentially improve public discourse on drug shortages and related public policy decisions.