Screening computed tomography†
Will it result in overdiagnosis of renal carcinoma?
The views expressed in this article are those of the authors and not necessarily the Robert Wood Johnson Foundation.
Despite uncertain benefits and harms, screening computed tomography (CT) is being marketed to consumers in the U.S. One plausible harm is the detection and treatment of renal carcinoma cases that otherwise would have remained clinically silent during the patient's lifetime.
After estimating the prevalence of preclinical renal carcinoma using meta-analysis of five series of asymptomatic, middle-aged Americans who received CT screening, the authors divided the prevalence by U.S. incidence rates of clinical renal carcinoma among persons of similar age. This calculation would estimate the mean duration of the detectable preclinical period (the “sojourn time”) of renal carcinoma if the incidence of preclinical and clinical renal carcinoma were equivalent.
The 5 series included 16,174 screenees (mean age range, 58–64 years; 61% male). The prevalence of asymptomatic renal carcinoma ranged from 0.11% to 0.76%; the pooled prevalence was 0.21% (95% confidence interval, 0.14–0.28%). The estimated mean sojourn time for renal carcinoma was between 3.7 years and 5.8 years among middle-aged Americans.
Because most renal carcinomas grow slowly during the preclinical period, the authors' estimated mean sojourn time did not seem unduly long. Therefore, the incidence rate of clinical renal carcinoma most likely is a reasonable surrogate for the incidence rate of preclinical renal carcinoma, implying that most renal carcinomas detected by CT screening among middle-aged Americans are likely to progress to clinical diagnosis. Cancer 2004;100:986–90. © 2004 American Cancer Society.
The objective of screening is to identify disease in its preclinical state when treatment is relatively more effective. If the duration of the preclinical disease state is very short, then effective screening may be impossible, because preclinical detection of a majority of cases would require an impracticably short screening interval. Conversely, if the duration of the preclinical disease state is very long or indefinite, then screening may detect conditions that would not progress to symptoms during the lifetime of some patients. Early detection of disease that ultimately would not present clinically in some persons is a plausible harm of cancer screening, because it would lead to treatment, often invasive, of no clinical benefit.1 To date, evidence suggests that “overdiagnosis” of this type is present in screening for prostate carcinoma,2 breast carcinoma,3 and lung carcinoma4 and for neuroblastoma.5
The advent of computed tomography (CT) screening raises the question of whether overdiagnosis may be clinically important for other solid organ malignancies. Many private radiologic corporations now market screening CT directly to consumers. Most companies offer “whole-body scans,” during which patients are scanned from the chest to the pelvis, although targeted scans of the lungs, heart, and colon also are available. During many CT screening examinations, images of the kidneys are obtained; thus, targeted CT screening often can function as a screening test for renal carcinoma.
We hypothesized that CT screening may be associated with overdiagnosis of renal carcinoma. To assess the likely extent of overdiagnosis of renal carcinoma, we first reviewed case series of CT screening to determine the frequency of detection of preclinical renal carcinoma. Then, under the assumption that all screen-detected tumors ultimately would progress to clinical diagnosis in the absence of screening, we divided the prevalence of preclinical renal carcinoma by incidence rates of renal carcinoma obtained from U.S. population-based cancer registries. If our assumption was valid, then this would estimate the mean duration of the preclinical period, also termed the “sojourn time.” We judged that overdiagnosis of renal carcinoma with CT screening could occur frequently if the estimated sojourn time were implausibly long, thereby implying that our underlying assumption regarding the inevitability of progression was incorrect.
MATERIALS AND METHODS
We sought to identify all published studies of CT screening that reported screening prevalence of renal carcinoma among asymptomatic adults in the U.S. After locating four studies from personal files,6–9 we hand searched the reference lists of each study to locate additional studies, but we found none. We then searched MEDLINE to identify articles similar to the four included studies, using the medical subject headings “tomography, X-ray computed” or “colonography, computed tomographic” and “mass screening.” One additional study was located in this search.10 We excluded studies that were performed outside the U.S. because we planned to compare reported screening prevalence with U.S. incidence rates of renal carcinoma.
We calculated the prevalence of preclinical renal carcinoma in each case series as well as the standard error of the prevalence estimate in each study.11 We estimated the pooled prevalence and its 95% confidence interval using meta-analysis of the individual study estimates and their standard errors. In the fixed-effects model, the pooled prevalence estimate is a weighted average of the individual study prevalences in which the weights are the inverse of the individual study variances.12 Because fixed-effects and random-effects models yielded identical results, and there was no large degree of statistical heterogeneity (P = 0.55), we only reported the results from the fixed-effects model. Analyses were conducted using the “meta” command in STATA statistical software (College Station, TX).
We estimated the mean sojourn time of renal cancer by calculating the ratio of the pooled prevalence of preclinical renal cancer across all five series and the U.S. incidence rates of carcinoma of the kidney and renal pelvis for age ranges that were similar to the mean ages of the study populations.13 This estimate is based on two assumptions: 1) the incidence rate of clinically diagnosed renal carcinoma approximates the incidence of preclinical renal carcinoma and 2) the incidence rate of renal carcinoma in the pooled study population is similar to incidence rate among a U.S. population of similar age with the same gender distribution. Incidence rates of carcinomas of the kidney and renal pelvis are compiled by federally supported regional cancer registries (the Surveillance, Epidemiology, and End Results Program).14
For the base case estimate, we used the following inputs: the pooled estimate of renal carcinoma prevalence and the incidence rate of carcinomas of the kidney and renal pelvis for U.S. residents ages 60–64 years adjusted for the gender distribution of the pooled study population. This choice was dictated by the age and gender distribution of participants in the included series (mean age, ≈ 60 years; 60.9% male). We then performed two-way sensitivity analyses: first, assuming the true prevalence of preclinical renal carcinoma was at the lower and upper bounds of the 95% confidence interval of the pooled prevalence estimate; and, second, assuming higher incidence rates of clinical disease (for U.S. adults ages 65–69 years and 70–74 years; adjusted for the gender distribution of the current study population).
We uncovered 5 series of screening CT that reported the prevalence of renal carcinoma, altogether including 16,174 adults in the U.S. (Table 1).6–10 In all series but one,7 the diagnoses of renal carcinoma were confirmed by medical record review. The largest series,6 which included 77% of the total individuals, reported the screening prevalence of renal carcinoma among self-referred patients receiving coronary artery disease screening with electron-beam CT. Another series reported results of screening “whole-body” CT among self-referred individuals.7 The remaining series were reports of clinical investigations of screening CT at the Mayo Clinic (Rochester, MN). By definition, all CT scans were performed for screening purposes, so no individuals had been referred for evaluation of clinical symptoms or signs of renal carcinoma.
Table 1. Case Series of Computed Tomography Screening That Reported the Prevalence of Preclinical Renal Carcinoma
|Mitchell et al., 20006||Self-referred screenees undergoing electron-beam CT screening for coronary artery disease (Dallas, TX; 64% men; mean age of cases, 58 yrs; age range, 34–74 yrs)||11,932||27||0.23||0.04|
|Brant-Zawadzki, 20027||Self-referred screenees for whole-body CT screening (Southern California; 48% men; age not reported)||1777||2||0.11||0.08|
|Swensen et al., 20028||Current or former smokers in lung carcinoma screening trial (Mayo Clinic, Rochester, MN; mean age, 59 yrs; 52% men)||1520||4||0.26||0.13|
|Gluecker et al., 200310||Case series of CT screening for colon carcinoma (colonography) among screenees at increased risk of colon carcinoma (Mayo Clinic, Rochester, MN; mean age, 64 yrs; 63% men)||681||2||0.29||0.21|
|Hara et al., 20009||Case series of CT screening for colon carcinoma (Mayo Clinic, Rochester, MN; mean age, 64 yrs; 55% men)||264||2||0.76||0.53|
The prevalence of preclinical renal carcinoma ranged from 0.11–0.76% (Table 1). The pooled estimated prevalence of preclinical renal carcinoma was 0.21% (95% confidence interval, 0.14%, 0.28%), or 2.1 cases per 1000 individuals screened.
Based on the pooled estimated prevalence of 0.21%, the mean sojourn time of preclinical renal carcinoma was 3.7–5.8 years, depending on the assumed incidence rate of clinical disease (Table 2). In sensitivity analyses testing the effect of assuming lower or higher screening prevalence, the estimated sojourn time ranged from a minimum of 2.5 years to a maximum of 7.7 years.
Table 2. Sensitivity Analyses Testing the Effect on Estimated Sojourn Time of Varied Assumptions Regarding the Prevalence at Screening and Incidence Rates of Renal Carcinoma
Recent reports of case series of screening CT have allowed us to estimate the prevalence of preclinical renal carcinoma among middle-aged Americans. By dividing the screening prevalence estimates by the incidence rates of clinical disease as a surrogate for the incidence of preclinical disease, we have estimated the mean sojourn time of preclinical renal carcinoma to be between 3.7 years and 5.8 years among a study population with a mean age of approximately 60 years. In sensitivity analyses assuming lower and higher screening prevalence, the mean sojourn time of preclinical carcinoma ranged from 2.5–7.7 years. If many preclinical renal carcinoma cases did not progress to clinical diagnosis, then the incidence rate of preclinical renal carcinoma would exceed by far the incidence rate of clinical renal carcinoma, and the denominators in our calculations would be too low. Therefore, to judge the likely extent of overdiagnosis of renal carcinoma with CT screening, it is necessary to consider whether our estimates of the mean sojourn time are unduly high.
We believe that a mean sojourn time for renal carcinoma of approximately 2.5–7.7 years is plausible for a primary tumor with a typically slow growth rate compared with other human malignancies.15 In a case series of 40 incidentally detected primary solid renal tumors that were monitored with serial CT, most tumors grew very slowly during a mean follow-up of 3.25 years.16 Slow growth rates also were observed in two small series of Japanese patients with incidentally detected primary renal carcinoma.17, 18 Moreover, to our knowledge, no instances of regression of renal carcinoma have been observed, although each series was small. Because our estimates of the sojourn time are consistent with the relatively indolent growth of preclinical renal carcinoma, we believe that the assumption underlying our calculation—that the incidence rate of clinical disease reasonably approximates the incidence rate of preclinical disease—most likely is valid. In other words, our results support the hypothesis that many, if not most, asymptomatic renal carcinomas detected by CT screening among middle-aged Americans would eventually progress to clinical diagnosis in the absence of screening.
We have estimated the mean duration of the sojourn time of renal carcinoma but were unable to infer the distribution of sojourn times based on our calculations. Indeed, our estimated mean sojourn time is consistent with a distribution of sojourn times composed of many extremely indolent tumors as long as a similar proportion of aggressive tumors had extremely short sojourn times. Although the distribution of sojourn times for most cancers is unknown,19 the series described by Bosniak et al. sheds light on the probable sojourn time distribution of primary renal carcinoma.16 Indeed, the distribution of growth rates appeared to be approximately normal, with small fractions of tumors growing either rapidly or extremely slowly. Thus, although data are limited, some empiric evidence suggests that only a small minority of renal carcinomas exhibit extremely indolent growth and, thereby, have very prolonged sojourn times.
In our calculations, we also assumed that incidence rates of renal carcinoma among individuals seeking screening CT would be similar to incidence rates of populations in the U.S. of similar ages. If such individuals are at lower or higher risk of renal carcinoma compared with the U.S. population, then our estimates of the mean sojourn times may be underestimated or overestimated, respectively. Comprising 9% of our study population, the patients in one series8 all were current or former smokers and, thus, most likely were at an increased risk of renal carcinoma relative to the general population of the U.S. Nevertheless, we lack sufficient data to judge whether the patients in the five series, collectively, were at relatively high risk or low risk of renal carcinoma.
The 37 patients who were diagnosed with renal carcinoma in the current series were at risk of mortality from other causes after diagnosis. In a large enough series, a proportion of newly diagnosed patients undoubtedly would die from other causes shortly after screening and prior to the time that renal carcinoma would have been diagnosed in the absence of screening. If a high proportion of patients diagnosed by screening were to die of competing causes, then the incidence rate of preclinical disease would exceed substantially the incidence rate of clinically diagnosed cancer. However, our methods suggest that the incidence rate of clinical cancer approximates the incidence rate of preclinical cancer, although we acknowledge that it most likely underestimates the true incidence of preclinical cancer to some degree. Therefore, the results of the current study do not imply that overdiagnosis of renal carcinoma never occurs with CT screening; rather, they suggest that it is unlikely that the degree of overdiagnosis is very large among a middle-aged U.S. population.
We have pooled prevalence data regarding renal carcinoma from case series of CT screening with incidence rates of clinical renal carcinoma from regional cancer registries in the U.S. to estimate the mean duration of the preclinical period of renal carcinoma. Our estimate of the preclinical period of approximately 4–6 years appears plausible based on observed growth rates of incidentally detected renal carcinomas. Therefore, the incidence of clinical renal carcinoma most likely is a reasonable surrogate for the incidence of preclinical disease and many, if not most, preclinical renal carcinomas detected by CT screening among middle-aged Americans will progress to clinical symptoms and diagnosis in the absence of screening.