Article Text

Original article
Gastric adenocarcinoma screening and prevention in the era of new biomarker and endoscopic technologies: a cost-effectiveness analysis
1. Jennifer M Yeh1,
2. Chin Hur2,
3. Zachary Ward1,
4. Deborah Schrag3,
5. Sue J Goldie1
1. 1Center for Health Decision Science, Harvard T. H. Chan School of Public Health, Boston, Massachusetts, USA
2. 2Massachusetts General Hospital Institute for Technology Assessment, Boston, Massachusetts, USA
3. 3Dana-Farber Cancer Institute, Harvard Medical School, Boston, Massachusetts, USA
1. Correspondence to Dr Jennifer M Yeh, Center for Health Decision Science, Harvard T. H. Chan School of Public Health, 718 Huntington Avenue, Boston, MA 02115, USA; jyeh{at}hsph.harvard.edu

## Abstract

Objective To estimate the cost-effectiveness of noncardia gastric adenocarcinoma (NCGA) screening strategies based on new biomarker and endoscopic technologies.

Design Using an intestinal-type NCGA microsimulation model, we evaluated the following one-time screening strategies for US men: (1) serum pepsinogen to detect gastric atrophy (with endoscopic follow-up of positive screen results), (2) endoscopic screening to detect dysplasia and asymptomatic cancer (with endoscopic mucosal resection (EMR) treatment for detected lesions) and (3) Helicobacter pylori screening and treatment. Screening performance, treatment effectiveness, cancer and cost data were based on published literature and databases. Subgroups included current, former and never smokers. Outcomes included lifetime cancer risk and incremental cost-effectiveness ratios (ICERs), expressed as cost per quality-adjusted-life-year (QALY) gained.

### Natural history simulation model

As depicted in figure 1, the model simulates the development of intestinal-type NCGA through a series of precancerous lesions, which may progress to dysplasia and eventually invasive cancer. At the start of the simulation, 20-year-old individuals enter the model and are assigned a risk factor profile for H. pylori and smoking status. Based on epidemiologic data,28 we assumed that precancerous lesions were already present in a subset of 20 years olds, with a greater proportion among those infected with H. pylori. Based on monthly probabilities derived via model calibration (see online supplementary table S1),22 individuals transition among the health states and are followed throughout their lifetime.

Figure 1

Diagram of intestinal-type noncardia gastric adenocarcinoma (NCGA) natural history model. Intestinal-type NCGA develops through a series of precancerous health states as depicted. Each month, individuals face a risk of progression among the health states. Before invasive cancer develops, individuals can also regress to less advanced precancerous lesions. Individuals with preclinical (asymptomatic) cancer can remain asymptomatic or progress to symptomatic clinical cancer. Once individuals develop symptomatic cancer, they are assumed to receive treatment and do not progress to more advanced cancer states. All probabilities are constant, except for the age-specific transition from dysplasia to preclinical cancer. The model was programmed in the computer language C++.

We based H. pylori prevalence and smoking profiles for a 1961–1965 birth cohort (corresponding to 50-year-old men between calendar years 2011 and 2015) on National Health and Nutrition Examination Survey and National Health Interview Survey data.29–32 Specifically, we estimated that 31% of the birth cohort was infected with H. pylori and assumed the following: H. pylori infection is established by age 20 years,33 causes gastritis and increases the risk of atrophy34 and remains unchanged throughout one's lifetime.35 ,36 We assumed that an individual's smoking status may change over their lifetime and estimated that 23% of individuals at age 50 years were current smokers (a decline from a peak prevalence of nearly 40% at age 30 years).31 ,32 We assumed that smoking increases the risk of progression to intestinal metaplasia and dysplasia,11 ,12 and that the magnitude rises with smoking intensity (defined as <10, 10–19, ≥20 cigarettes per day). Upon quitting, individuals face former smoker-specific disease progression rates.37

For individuals who developed intestinal-type NCGA, stage-specific mortality rates were based on SEER estimates.24 Competing cause mortality was based on US birth cohort-specific life tables38 ,39 and adjusted for smoking intensity using published relative risk estimates.40

To assess the face validity and projective validity of the model, we compared model outputs to data not used for model parameterisation or calibration. Model estimates of the relative risk of intestinal-type NCGA associated with H. pylori infection (3.6) and smoking (1.6) were consistent with published estimates (95% CI 2.7 to 7.2 and 1.5 to 1.8, respectively).7 ,41 The proportion of all cancers occurring in H. pylori-positive individuals (60%) fell within the calculated population attributable fraction range for the cohort (38–65%).42 Modelled estimates for prevalence of precancerous lesions and 10-year cancer risk for individuals with dysplasia also approximated published estimates (see online supplementary materials for full details).15 ,43

### Strategies

Compared with no screening, we evaluated the following one-time screening strategies at age 50 years: (1) serum pepsinogen screening, (2) endoscopic screening and (3) H. pylori screening.

For serum pepsinogen screening, all individuals with a positive assay test result for atrophy (defined as serum pepsinogen I levels ≤70 µg/L together with pepsinogen I/II ratio ≤3.0) were followed up by endoscopy with seven to nine random biopsies of the gastric mucosa.44 All individuals with a positive endoscopy for dysplasia or asymptomatic localised cancer (detected either macroscopically via endoscopy or histologically based on gastric biopsies) underwent EMR treatment to remove lesions; those with a negative endoscopic result returned for a follow-up endoscopy in 10 years. Individuals with an initial negative serum pepsinogen test result received no further treatment or follow-up. As part of standard practice, we assumed that individuals would also be tested for H. pylori and all individuals who tested positive for the infection would receive standard triple therapy (20 mg omeprazole, 1 g amoxicillin, 500 mg clarithromycin, twice daily).

For endoscopic screening, individuals with a positive screen result (based on biopsy sampling) for dysplasia or asymptomatic localised cancer received EMR treatment; those with a negative biopsy result received no additional follow-up care.

For both the serum pepsinogen and endoscopic screening strategies, we assumed: (1) only asymptomatic localised cancers detected via endoscopy benefited from screening (ie, negligible survival benefit for detected regional and distant cancers), (2) all dysplastic lesions (detected macroscopically and/or histologically) and submucosal localised cancers (American Joint Committee on Cancer stage IA) were eligible for EMR treatment; all other localised tumours were surgically treated, and (3) all individuals treated with EMR returned for post-treatment surveillance via endoscopy for recurrence in 10 years.

For H. pylori screening, all individuals with a positive test result received standard 10-day triple therapy (described above).

We evaluated the screening strategies for the overall cohort and smoking subgroups based on smoking status at time of screening (never, current or former smokers).

### Clinical data

For each strategy, we estimated screening test characteristics, complication rates and treatment effectiveness based on data from published clinical studies (table 1).18 ,20 ,24 ,45–64 For the serum pepsinogen test, sensitivity and specificity were based on the ability to detect atrophy (with and without more advanced precancerous lesions), with dysplasia and asymptomatic cancerous lesions identified via subsequent endoscopy and random biopsy sampling. All endoscopic procedures, including EMR, were associated with a risk of severe bleeding or perforation requiring surgery.50 ,52 After EMR treatment, individuals faced a risk of recurrence from incomplete resections and, for cancerous lesions, metachronous lesions (which we assumed stemmed from undetected dysplasia).51 ,52 ,57 For H. pylori treatment, we assumed standard triple therapy reduced the risk of progressing from gastritis to atrophy to H. pylori-negative rates20 with 80% efficacy.56

Table 1

Select model parameters: base case value and plausible range

To reflect quality of life, we used sex- and age-specific population-based weights58 and disease-specific weights.59 We also assumed that endoscopic procedures and surgical procedures were associated with a 50% utility reduction for 1 day and 2 weeks, respectively.

### Cost data

For each strategy, we estimated direct medical costs based on 2012 Medicare reimbursement rates (table 1). Costs included physician costs, pathologist costs (for biopsy evaluation), and facilities and/or hospitalisation costs associated with endoscopic or surgical procedures.60 We used phase-specific treatment costs for gastric cancer.62 Drug costs were based on median Wholesale Acquisition Cost among leading manufacturers.61 Indirect patient costs were based on time lost from work,63 including phase-specific time costs for cancer treatment.64

### Sensitivity and uncertainty analysis

We conducted sensitivity analyses on key variables to explore how results varied across plausible ranges established from published studies. To reflect the impact of uncertainty surrounding disease natural history on results, we conducted analyses with a subset of 50 good-fitting natural history parameters identified via model calibration (see online supplementary table S1) and report the range across all parameter sets for all model outcomes. In addition, we conducted a probabilistic sensitivity analysis using 1000 second-order Monte Carlo simulations, in which key model parameters, including natural history parameter sets, were simultaneously varied.

## Results

### Clinical benefits

#### General population

For a hypothetical cohort of 20-year-old men, the modelled lifetime risk of intestinal-type NCGA was 0.24% (table 2). The relative reduction in intestinal-type NCGA lifetime risk was 26.4% with serum pepsinogen screening (range 22.3–34.5%), 21.2% with endoscopic-based screening (range 17.0–30.1%) and 0.2% with H. pylori screening at age 50 years (range 0.0–1.2%) (figure 2). The gain in life expectancy was greatest for serum pepsinogen screening (2.7 days; range 2.4–4.2 days) compared with endoscopy with EMR (2.4 days; range 2.1–3.9 days) and H. pylori screening and treatment (0.01 days; range 0.00–0.07 days) (table 2). Among individuals with a positive serum pepsinogen result, the life expectancy gain was 1.2 months, and among those with a positive follow-up endoscopy for dysplasia or asymptomatic cancer, 1.2 years.

Table 2

Clinical and economic model outcomes associated with noncardia gastric adenocarcinoma (NCGA) screening strategies for the overall cohort and smoking subgroups (per-person averages)

Figure 2

Relative reduction in lifetime intestinal-type noncardia gastric adenocarcinoma (NCGA) risk among the overall cohort and smoking subgroups. Serum pepsinogen screening was associated with the greatest relative reduction in lifetime intestinal-type NCGA risk compared to no screening. The reduction associated with Helicobacter pylori screening was the smallest (<0.2%). Results were similar for the overall cohort (black bars) and smoking subgroups, including never smokers (white bars), current smokers (dark grey bars) and former smokers (light grey bars). Error bars depict the range among the subset of 50 randomly selected good-fitting natural history parameter sets. A positive serum pepsinogen screen was defined as: pepsinogen I levels ≤70 µg/L and pepsinogen I/II ratio ≤3.0.

Among a cohort of 10 million 20-year-old men, the model estimated that serum pepsinogen screening would prevent 5126 (range 4687–8316), or 27.0% (range 23.2–34.7%), of the projected 19 014 intestinal-type NCGA deaths (range 16 576–26 255) (table 3). The estimated number needed to screen (NNS) to prevent one intestinal-type NCGA death was 1813 (range 1117–1982). The estimated number of endoscopies needed was 295 (range 216–378). Table 3 depicts additional results.

Table 3

Estimated number of intestinal-type noncardia gastric adenocarcinoma (NCGA) deaths, screening-related deaths and net number of deaths averted associated with NCGA screening strategies for the overall cohort and smoking subgroups

#### Current or former smokers

The relative reduction in lifetime intestinal-type NCGA risk was greatest among current smokers (figure 2). Targeting current smokers at age 50 years reduced the lifetime risk (0.35%) by 30.8% with serum pepsinogen screening (range 27.0–38.5%), 25.5% with endoscopy and EMR (range 21.5–34.8%) and 0.1% with H. pylori screening and treatment (range 0.0–1.0%). The number of days gained for each strategy was higher for current smokers (table 2). Current smokers with a positive serum pepsinogen screen test result also had a greater gain in life expectancy compared to never or former smokers (1.4 months vs 1.1–1.2 months, respectively).

Among the approximately 2.10 million individuals who were current smokers at age 50 years, serum pepsinogen screening would prevent 1810 (range 1602–2476), or 31.4% (range 27.8–38.5%), of the projected 5758 intestinal-type NCGA deaths (range 4836–7344). The percent reduction in number of intestinal-type NCGA deaths prevented was similar for never and former smokers, reflecting the reduced risk of progressing to invasive cancer associated with smoking cessation. The model estimated that approximately 1157 current smokers (range 846–1307) would need to be screened to prevent one intestinal-type NCGA death. The corresponding number of endoscopies needed was 218 (range 174–297).

### Sensitivity and uncertainty analysis

For the overall cohort, results for the serum pepsinogen screening strategy were most sensitive to H. pylori prevalence, screen age, serum pepsinogen test sensitivity, and costs associated with endoscopic follow-up (figure 3). Results were moderately sensitive to serum pepsinogen screening costs and test specificity. Results remained largely unchanged over the plausible range for endoscopic sensitivity for dysplastic and cancerous lesions, EMR treatment effectiveness and complication risks, and proportion of EMR-eligible localised cancers.

Figure 3

Tornado diagram on one-way sensitivity analysis for serum pepsinogen screening strategy: select model parameters. Based on one-way sensitivity analyses, this figure depicts the relative influence of select model parameters on results for serum pepsinogen screening for the overall cohort. The x-axis shows the effect of changes in selected variables on the incremental cost-effectiveness ratio (ICER) for serum pepsinogen screening at age 50 years (compared to no assessment). The y-axis shows selected model parameters, with the base case value and range used in the sensitivity analysis shown in parentheses. The shaded bars indicate the variation in the ICER caused by changes in the value of the indicated variable while all other variables were held constant. The dotted vertical black line indicates the ICER for the base case. The solid vertical grey line represents the commonly used $100 000 per quality-adjusted-life-year cost-effectiveness threshold. *The first number in the range indicates value yielding the lowest ICER; the second indicates value yielding the highest ICER. H. pylori, Helicobacter pylori; EMR, endoscopic mucosal resection. We conducted scenario analyses for the overall cohort to explore alternative model assumptions. In our base case, we used estimates for surgical mortality risks among asymptomatic individuals in good health; if age-specific risks for symptomatic patients undergoing surgery were used instead,65 the ICER for serum pepsinogen screening increased to$128 400 per QALY gained. If 5% of endoscopic procedures required hospitalisation (related to complications or incidental findings that required follow-up care), the ICER increased to $120 600 per QALY gained. Similarly, if follow-up endoscopic surveillance was based on only macroscopic findings (ie, no gastric biopsies were taken), both the reduction in cancer risk (21% vs 26% in base case) and attractiveness of serum pepsinogen screening ($130 000 vs $105 400 per QALY gained in base case) declined. If we assumed that after EMR treatment, individuals still harboured intestinal metaplastic lesions (which could progress to invasive cancer), serum pepsinogen screening was also less attractive (ICER=$116 000 per QALY gained). For all these scenarios, ICERs remained less than $93 000 per QALY gained for current smokers. To assess the impact of H. pylori prevalence, we determined the threshold value needed for serum pepsinogen screening to be considered cost-effective. At a$100 000 per QALY gained threshold, 40% of the cohort would need to be H. pylori infected (base case=31%) (see online supplementary figure S1). For current smokers, screening was considered cost-effective at nearly all prevalence levels. In contrast, for never smokers, the ICER exceeded the $100 000 per QALY threshold at all prevalence levels. Two-way threshold analysis on serum pepsinogen test characteristics similarly found that the range of possible values for which serum pepsinogen would be preferred was much broader for current smokers compared to the other subgroups (figure 4). For current smokers, if sensitivity was greater than 60%, serum pepsinogen screening was the preferred strategy as long as test specificity was greater than 94%. For never smokers, serum pepsinogen screening was preferred only if the test had nearly perfect performance. Figure 4 Two-way threshold analysis on serum pepsinogen test characteristics for the overall cohort and smoking subgroups. The preferred strategy based on serum pepsinogen screening test sensitivity and specificity are shown for the overall cohort (Panel A) and smoking subgroups (never smokers (Panel B), current smokers (Panel C), and former smokers (Panel D)). In each panel, the shaded grey region indicates the range of values over which serum pepsinogen screening would be considered the preferred strategy at a cost-effectiveness threshold of$100 000 per quality-adjusted-life-year gained. For example, for never smokers, serum pepsinogen screening would be the preferred strategy only with nearly perfect test sensitivity and specificity. For all possible test characteristic values depicted, serum pepsinogen screening dominated all other screening strategies, in that it was either less costly and more effective (endoscopic screening) or more effective and more cost-effective (Helicobacter pylori screening). A positive serum pepsinogen screen was defined as: pepsinogen I levels ≤70 µg/L and pepsinogen I/II ratio ≤3.0.

For the overall cohort, probabilistic sensitivity analysis suggested that at a cost-effectiveness threshold of $100 000 per QALY gained, the probability that serum pepsinogen screening was the preferred strategy was 0.47 (table 2 and figure 5). The probability was 0.97 for current smokers and 0.85 for former smokers. Figure 5 Cost-effectiveness acceptability curves for the overall cohort and smoking subgroups. To illustrate the uncertainty surrounding incremental cost-effectiveness ratio estimates, the cost-effectiveness acceptability curves depict the probability that a given strategy is the preferred strategy across a range of cost-effectiveness ratios. Results are depicted for the overall cohort (black solid lines) and subgroups, including never smokers (grey dotted line), current smokers (grey long dashed line), and former smokers (grey short dashed line). Results are based on 1000 second-order Monte Caro simulations in which model variables were simultaneously varied. The solid black vertical line indicates the$100 000 per quality-adjusted-life-year willingness-to-pay threshold commonly used as a benchmark in the US.

## Discussion

Although intestinal-type NCGA incidence has declined over the past century, disease risk is largely determined by H. pylori infection acquired in childhood and the number of cases is projected to remain considerable for decades.22 To provide insight into this important public health and clinical problem and explore options for secondary prevention, we employed a model-based approach to estimate the comparative benefits and cost-effectiveness associated with several screening strategies. Our findings suggest that although a one-time serum pepsinogen screening (with endoscopic follow-up and EMR treatment if needed) at age 50 years can prevent as many as one in four intestinal-type NCGAs among US men, general population-wide screening is unlikely to be a high-value strategy for improving cancer outcomes. However, screening targeted to current smokers who are at elevated risk for premature death66 may be an effective and cost-effective strategy to reduce NCGA mortality.

Our study is the first simulation model-based analysis to evaluate the clinical benefits and economic consequences of serum pepsinogen screening in the US. Previous model-based studies have focused on high-risk populations in Asia,67 or estimated the short-term economics of serum pepsinogen testing as a follow-up strategy for individuals diagnosed with atrophy or intestinal metaplasia.68 Neither study estimated the clinical benefits associated with serum pepsinogen screening in terms of a reduction in cancer risk, or provided estimates for smoking subgroups which can be used as the basis for targeting screening efforts. Consistent with published studies (see online supplementary table S2),45 ,69–74 our model-based estimates of serum pepsinogen screening performance underscore the potential usefulness of the test to detect and distinguish individuals at higher risk for developing cancer from those at lower risk.75 ,76 Furthermore, our estimates of the NNS to prevent one intestinal-type NCGA death (1157 US male smokers) suggest that serum pepsinogen screening may have similar benefits to mammography screening among 50–59-year-old women (NNS=1339 to prevent 1 breast cancer death), albeit smaller benefits than low-dose computed tomography (NNS=320 to prevent 1 lung cancer death)77–79 or flexible sigmoidoscopy (NNS=871 to prevent 1 colorectal cancer death).80 However, our findings should be cautiously considered given the notable uncertainty surrounding serum pepsinogen test performance,17 limited evidence in low-risk populations,18 and concerns surrounding the translation of clinical findings in high-risk populations to low-risk populations.81 As better data become available, our model can be refined and recalibrated to reflect these data, and as such, can serve as an iterative tool to provide updated assessments of the likely health and economic outcomes associated with secondary gastric cancer control efforts.

The serum pepsinogen test has also been proposed as the basis for screening for intestinal-type NCGA itself (in contrast to identifying individuals with atrophy who are at elevated risk for NCGA as in our analysis). Our model found, however, that such a strategy, with a 77% sensitivity and 73% specificity for dysplastic and cancerous lesions,17 would not be cost-effective in the US, even among current smokers (see online supplementary table S3). While such a test would lead to similar reductions in cancer risk as atrophy screening, nearly 30% as opposed to 2% would have a false positive test and receive treatment unnecessarily leading to higher costs and possible harm without a commensurate gain in benefits. Similar to our threshold analyses on screening test characteristics (figure 4), these results highlight the importance of accurately detecting the absence of atrophy or precancerous lesions for any serum pepsinogen test-based screening strategy.

Previous studies have concluded H. pylori screening is cost-effective in the US.21 ,82 Our findings provide updated estimates of the cost-effectiveness of population-based H. pylori screening based on randomised trial evidence that only individuals without existing precancerous lesions benefit from H. pylori treatment.20 Under this assumption, we found that targeting screening to 20-year-old individuals (who are less likely to have precancerous lesions) may be more effective in reducing cancer risk (1.6% vs 0.2%). However, even with the greater benefit, the strategy would remain unattractive compared to no screening (ICER=$2.7M per QALY) and dominated by serum pepsinogen screening. Recent results from another randomised trial in China suggest that all individuals, regardless of the presence of advanced lesions, may benefit from H. pylori treatment,83 potentially as a result of eradicating non-H. pylori bacteria that influence the later stages of gastric carcinogenesis.84 If we assumed that the risk of dysplasia was reduced by 50% among all treated individuals, H. pylori screening was indeed more attractive compared to no screening (21% reduction in cancer risk at an ICER of$85 000 per QALY). Yet, H. pylori screening was still dominated by serum pepsinogen screening as the reduction in cancer risk was also greater for serum pepsinogen screening (44%) and at a more favourable ICER ($70 700 per QALY). As such, despite the considerable uncertainty in H. pylori treatment effectiveness, our findings suggest that serum pepsinogen is likely to be a more effective and cost-effective NCGA screening strategy in the US. Limitations to our study include using data from multiple sources with varying study designs. We conducted extensive probabilistic sensitivity analyses to account for the uncertainty in variables and assumptions, including disease natural history. We focused on only men, and made the simplifying assumption that the prevalence of H. pylori and smoking were independent because data regarding interactions are not available. We also only focused on one gastric cancer subtype. If we assumed that diffuse and other noncardia tumours (detectable for 24 months on average before becoming clinically symptomatic) were also detected via follow-up endoscopy for a positive serum pepsinogen screen, results were largely unchanged (ICER=$104 100 vs $105 400 in the base case). This was consistent with our finding that the majority of serum pepsinogen screening benefit was derived from the detection and removal of dysplastic lesions before they progressed to invasive cancer. We found however, that serum pepsinogen screening was less attractive if treated individuals still harboured intestinal metaplastic lesions (common in settings where H. pylori-related atrophy is frequently multifocal) or if endoscopic sensitivity for dysplasia was considerably lower. However, as long as sensitivity was greater than 60% (base case=81%), serum pepsinogen screening remained attractive for current smokers (ICER=$99 400).

Notably, we based estimates of serum pepsinogen test performance and EMR treatment effectiveness on clinical studies from Japan given the limited data in Western populations. EMR is relative new, availability of and expertise with EMR technology is limited; additional training (and resources) will be needed to realise the projected screening benefits. Not all biopsy-detected dysplasia may be macroscopically visible and therefore, eligible for EMR. We found however that even if the large majority of individuals with dysplasia (60–70%) would require annual or biannual endoscopic surveillance before undergoing EMR treatment, results were largely unchanged. We also did not include the impact of endoscopy-related incidental findings and their downstream effects in our analysis; further analysis of their long-term effects is needed. Proton-pump inhibitors, widely used in the general population, may reduce serum pepsinogen test sensitivity by altering intragastric acidity and biomarker levels.85 Sensitivity analyses found that even if sensitivity fell to 60% (base case=71%), as long as test specificity was greater than 94%, the ICER remained attractive for current smokers.

Last, our model focused on NCGA screening in the US. Estimates of the clinical benefits and cost-effectiveness associated with screening strategies will vary in high-risk countries, such as Japan, where risk factor prevalence and influence on the multifactorial aetiology of gastric carcinogenesis may differ. As the projective validity of our model was consistent with data on precancerous lesions prevalence and cancer risk from the Netherlands, our findings are likely generalisable to this setting and other low-risk European countries with similar H. pylori and smoking profiles.86

Our model-based findings suggest that serum pepsinogen screening to reduce NCGA risk is not warranted for the general population. However, targeting high-risk smokers for screening may be an effective and cost-effective strategy to reduce intestinal-type NCGA mortality. Further, the marginal benefits associated with H. pylori screening, even among high risk subgroups, underscore the need for future clinical studies on alternative secondary gastric cancer control strategies, including serum pepsinogen screening, to improve cancer outcomes and overall survival.

• ## Supplementary Data

This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.

Files in this Data Supplement:

## Footnotes

• Contributors JMY, CH, ZW, DS, SJG conceived and designed the study; analysed and interpreted the data; drafted and critically revised the manuscript for important intellectual content; and approved the version to be published. All authors had full access to all the data and take responsibility for the integrity of the data and the accuracy of the data analysis. JMY is the guarantor.

• Funding JMY was supported by the National Cancer Institute grant K07CA143044.

• Competing interests None.

• Provenance and peer review Not commissioned; externally peer reviewed.

## Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.