Overdiagnosis Bias Research Articles

Invited commentary

Invited commentary

Lung cancer screening with helical computed tomography in older adult smokers: a decision and cost-effectiveness analysis.

Encouraged by direct-to-consumer marketing, smokers and their physicians are contemplating lung cancer screening with a promising but unproven imaging procedure, helical computed tomography (CT). To estimate the potential benefits, harms, and cost-effectiveness of lung cancer screening with helical CT in various efficacy scenarios. Using a computer-simulated model, we compared annual helical CT screening to no screening for hypothetical cohorts of 100 000 current, quitting, and former heavy smokers, aged 60 years, of whom 55% were men. We simulated efficacy by changing the clinical stage distribution of lung cancers so that the screened group would have fewer advanced-stage cancers and more localized-stage cancers than the nonscreened group (ie, a stage shift). Our model incorporated known biases in screening programs such as lead time, length, and overdiagnosis bias. We measured the benefits of screening by comparing the absolute and relative difference in lung cancer-specific deaths. We measured harms by the number of false-positive invasive tests or surgeries per 100 000 and incremental cost-effectiveness in US dollars per quality-adjusted life-year (QALY) gained. Over a 20-year period, assuming a 50% stage shift, the current heavy smoker cohort had 553 fewer lung cancer deaths (13% lung cancer-specific mortality reduction) and 1186 false-positive invasive procedures per 100 000 persons. The incremental cost-effectiveness for current smokers was $116 300 per QALY gained. For quitting and former smokers, the incremental cost-effectiveness was $558 600 and $2 322 700 per QALY gained, respectively. Other than the degree of stage shift, the most influential parameters were adherence to screening, degree of length or overdiagnosis bias in the first year of screening, quality of life of persons with screen-detected localized lung cancers, cost of helical CT, and anxiety about indeterminate nodule diagnoses. In 1-way sensitivity analyses, none of these parameters was sufficient to make screening highly cost-effective for any of the cohorts. In multiway sensitivity analyses, a program screening current smokers was $42 500 per QALY gained if extremely favorable estimates were used for all of the influential parameters simultaneously. Even if efficacy is eventually proven, screening must overcome multiple additional barriers to be highly cost-effective. Given the current uncertainty of benefits, the harms from invasive testing, and the high costs associated with screening, direct-to-consumer marketing of helical CT is not advisable.

Is Autofluorescence Bronchoscopy Needed to Diagnose Early Bronchogenic Carcinoma?

Is Autofluorescence Bronchoscopy Needed to Diagnose Early Bronchogenic Carcinoma?

Natural history of pure ground-glass opacity after long-term follow-up of more than 2 years

Background. Pure ground-glass opacity (PGGO) is a new entity that has been clearly defined on high-resolution computed tomography (CT) during the last half decade. It is important to investigate the natural history of PGGO through long-term observation for the management of this new entity. Methods. We investigated 19 patients with PGGO(s) defined on high-resolution computed tomography and retained as PGGO for more than 2 years. The PGGOs of 11 patients were detected at annual mass screening by low-radiation–dose CT (low-dose CT), 7 at follow-up CT after cancer resection, and 1 incidentally on CT. After long-term observation, 10 of 19 patients underwent operation and 9 are currently being followed-up with CT. Their growth characteristics and histologic findings are reported. Results. The median follow-up period was 32 months, ranging from 24 to 124 months. The sizes of PGGOs at the time of discovery were 4 to 18 mm in largest diameter (average 8.6 mm). During follow-up, the size of PGGO showed no change in 8 patients, increased slightly (up to 5 mm) in 6 patients, and increased by more than 5 mm in 5 patients. Ten patients had limited resection (segmentectomy or wide wedge resection) with negative surgical margin by intraoperative lavage cytology of the resection margin of the lung. Of them, 5 patients had adenocarcinoma, 3 pulmonary lymphoproliferative disorder, and 1 each atypical adenomatous hyperplasia and focal fibrosis. There was no clear tendency between the degree of size change and histology. In all but 1 of 9 patients with follow-up only, the PGGOs showed either no change or only a slight increase within 5 mm in largest diameter. Conclusions. These data suggest that some PGGOs will never progress to clinical disease and would be included in the category of overdiagnosis bias. However, a prior history of lung cancer should significantly raise the index of suspicion, as 4 of 5 proven cancer cases in this small series fell into that category. Because of the difficulties of preoperative and intraoperative histodiagnosis of PGGO, minimally invasive surgery may be appropriate from the viewpoints of both diagnosis and curability.

Identification of lung cancer by autopsy: an indicator for clinical underdiagnosis

It is with great interest we read the article by S. Dammas et al. on ‘Identification of small lung nodules at autopsy: implications for lung cancer screening and overdiagnosis bias’, which appeared in the July 2001 issue of Lung Cancer [ [1] Dammas S. Patz E.F. Goodman P.C. Identification of small lung nodules at autopsy: implications for lung cancer screening and overdiagnosis bias. Lung Cancer. 2001; 33: 11-16 Abstract Full Text Full Text PDF PubMed Scopus (42) Google Scholar ]. The authors conclude that autopsies do not identify all small pulmonary nodules found at CT and that the true incidence of small clinically insignificant and thus, undiagnosed lung cancer (LC) is unknown. They further state that overdiagnosis of LC may be more important than previously recognised. These conclusions deserve some comment.

Promoting Ethical and Objective Practice in the Medicolegal Arena of Disability Evaluation

As providers of medical information and testimony, clinicians have ultimate responsibility for ethical conduct as it relates to this information. The authors offer the following recommendations for enhancing ethical relationships between expert clinicians and the courts. 1. Avoid or resist attorney efforts at enticement into joining the attorney-client team. Such compromises of scientific boundaries and ethical principles exist on a continuum ranging from standard attorney-client advocacy at the beginning of the expert consultation phase (e.g., promotional information at the forefront of retaining an expert, with either provision of selective or incomplete records or less than enthusiastic efforts to produce all records) and extending to completion of evaluation, when requests for changes in reports and documentation might be made. 2. Respect role boundaries and do not mix conflicting roles. Remember that the treating doctor possesses a bond with the patient but does not as a rule obtain complete preinjury and postinjury information in the context of assessing causality and apportionment. In contrast, the expert witness must conduct a thorough and multifaceted case analysis sans the physician-patient relationship in order to facilitate objectivity and allow optimum diagnostic formulations. Finally, the trial consultant's function in this adversarial process is to assist with critically scrutinizing and attacking positions of experts for the opposing side. These roles all represent inherently different interests, and mixing them can only reduce objectivity. 3. Insist on adequate time for thorough record review, evaluation, and report generation. Also insist on sufficient time and preparation for deposition and court appearances. 4. Work at building a reputation for general objectivity, reliance on multiple data sources, reaching opinions only after reviewing complete information from both sides, and completing the evaluation. 5. Spend a good amount of time actually treating the patient population being examined or being offered testimony about. This treatment should be current and should be of a similar frequency to treating practitioner specialists. Be able to discuss relevant research and scientific methodology issues competently and without notes. 6. Arrive at opinions only after reviewing all of the evidence from both sides of the adversarial fence, employing multiple data sources, completing the evaluation, and interpreting data within the full context of comprehensive historical, behavioral observation, and contextual information. Being otherwise favorable to retaining attorney interests suggests endorsement of "opinion prostitute," "scientific perjurer," or "hired gun" status. The only way a practitioner can reduce the likelihood of facing an "opinion prostitute" on the opposing side in future cases is to insist on establishing and maintaining a reputation for scientific objectivity. 7. Balance cases from plaintiff and defense attorneys. Predilection for one side or the other suggests bias and sets up predisposition to nonobjectivity. For example, a preponderance of plaintiff work suggests an overdiagnosis or uncritical sympathy bias, whereas a ratio that favors hiring by the defense suggests an underdiagnosis or skepticism bias. Perhaps Brodsky's suggested cut-off ratio of .8 for favorability findings would represent an initial cutoff for defense versus plaintiff ratio. That is, experts should do at least 20% work for the opposite side of the current case being represented. Further, it might be a reasonable expectation that data on these ratios be collected as an important method for ensuring objective opinions. 8. Ensure against excessive favorability to the side of the retaining attorney or firm. Objectivity demands that scientific opinions not be influenced by the position of the legal advocate. Importantly, Brodsky recommends using a ratio of .8 as a cut-off for detecting excessive bias. That is, practitioners should possess prerequisite objectivity to disagree with the referring attorney at least 20% of the time. We suggest that a more useful cut-off would be .75, where experts are expected to generate findings that do not support the referring attorney's position at least 25% of the time. 9. Never arrive at opinions that are inconsistent with plaintiff records, examination data, test data, behavioral presentation, and so forth, especially when such opinions are favorable to the side of the retaining attorney firm. Instead, use the following recommendations. (ABSTRACT TRUNCATED)

Identification of small lung nodules at autopsy: implications for lung cancer screening and overdiagnosis bias

Purpose: Unsuspected cases of lung cancer are reported to be uncommon in autopsy series, and these data have been used to suggest that indolent tumors are rare and that overdiagnosis bias is not an important factor in lung cancer screening. The purpose of this study was to determine if a retrospective autopsy review is indeed accurate in identifying all small lung nodules on CT, and thus provide a true estimate of unsuspected lung tumors. Materials and Methods: We identified all 1047 patients who had an autopsy at our institution from 1994 to 1998. We then reviewed the patients radiology records and found 187 patients with a thoracic CT within 2 months of the postmortem examination. All 187 CT reports were reviewed in order to identify patients with at least one pulmonary nodule. CT studies with reports that described a nodule(s) were then re-reviewed to confirm presence and location of the nodule(s). The CT findings were than compared to the autopsy report to determine if the postmortem examination indeed found the nodule(s). Results: 28 autopsy patients had at least one pulmonary nodule identified on their thoracic CT no more than 2 months before death. Nineteen patients (68%) had nodule(s) recorded on the autopsy report, two (∼10%) of which proved to have undiagnosed squamous cell carcinoma. Nine patients (22%) had no mention of pulmonary nodules seen on the CT recorded on their autopsy report. Conclusions: This study suggests autopsies do not identify all small pulmonary nodules found at CT. The true incidence of clinically insignificant lung cancer is thus uncertain, and overdiagnosis bias in lung cancer screening may be more important than previously recognized.

The risks of screening: data from the Nottingham randomised controlled trial of faecal occult blood screening for colorectal cancer

AIMSTo determine the harm that ensues from faecal occult blood (FOB) screening for colorectal cancer.METHODS150 251 people were randomly allocated either to receive biennial Haemoccult FOB tests (n =75 253)...

Advances in radiology and the real versus apparent effects of early diagnosis

Over the last two decades, technological advances in radiology have revolutionized the practice of medicine. Although the potential benefits of these advances are well recognized, the potential harms are not. This paper describes how early diagnosis can cause overestimations of disease prevalence and the effectiveness of intervention. The paper begins by demonstrating how the observed prevalence of disease increases with the sensitivity of the test and by explaining the concept of pseudodisease. Next, the paper explains how lead time bias, length bias and overdiagnosis bias cause overestimations of the effectiveness of earlier diagnosis and associated treatments. These biases pertain to both the detection and staging of disease. In addition, the paper explains how these overestimations of disease prevalence and the effectiveness of intervention can initiate a cycle of increasing testing and treatment, which may eventually cause more harm than benefit. Finally, randomized clinical trials and decision analysis are discussed in the context of evaluating new testing strategies.

Ductal Carcinoma In Situ: The Impact of Screening on Clinical Presentation and Pathologic Features

: It has been suggested that mammographic screening may produce an overdiagnosis bias leading to unnecessary treatment for inconsequential ductal-carcinomain-situ lesions. To assess the malignant potential of ductal carcinoma in situ, we compared pathologic features and presentation of ductal carcinoma in situ in patients treated before (1969–1985) and after (1986–1990) the intensified use of screening at our institution. We reviewed 204 cases of ductal carcinoma in situ in patients treated between 1969 and 1990. Histologic slides were available for 145 patients. Pathologic features evaluated included histologic subtype and nuclear grade. When the prescreening era was compared with the postscreening era, the incidence of palpable mass decreased from 54% to 12%, ductal discharge decreased from 14% to 3%, Paget's disease decreased from 10% to 3%, and mammographically detected lesions increased from 19% to 80%. The incidence of comedo subtype increased from 14% to 36% (p = 0.0195). Seven percent of palpable masses and 38% of mammographically detected ductal-carcinoma-in-situ lesions were of the comedo subtype, which presented almost exclusively as a mammographic abnormality (p < 0.0015). Grade 1 lesions decreased from 38% to 28%, and grade 3 lesions increased from 24% to 33%. Palpable masses tended to be lower grades and mammographically detected lesions tended to be higher grades. The malignant potential of mammographically detected lesions as determined by grade and histologic subtype is greater than that of symptomatic ductal carcinoma in situ. Mode of presentation does not appear to be a useful prognostic factor. Mammographically detected ductal carcinoma in situ is not an inconsequential finding and should not be treated as such.

Measuring Effectiveness of Lung Cancer Screening: From Consensus to Controversy and Back

While intense controversy exists regarding screening for breast, colorectal, and prostate cancer, a consensus exists regarding lung cancer screening. All organizations recommend against any efforts to detect early lung cancer because each of four randomized controlled trials (RCTs) has failed to demonstrate a significant reduction in lung cancer mortality as a result of screening. Disease-specific mortality is assumed to represent the best measure of screening effectiveness in RCTs, because it is not subject to confounding by lead time, length, or overdiagnosis biases. However, the effects of these biases are predictable, so accurate assessments of the degree of confounding by these biases can be made. Moreover, the ability of mortality to accurately reflect cancer death rates depends on the ability of randomization to create experimental and control populations that have an equal risk of dying of the disease under study, except insofar as early detection may reduce that risk. Because the majority of participants in screening trials never develop the disease under investigation, small absolute differences in disease risk between groups often persist despite randomization, and such differences translate into much larger proportional differences in the size of subgroups at risk for disease-specific mortality. This effect confounds the ability of disease-specific mortality to accurately measure screening effectiveness. A total of 18 RCTs have been conducted to evaluate screening for breast, colorectal, and lung cancer. In the only two RCTs that reported a significant mortality reduction for screening mammography in breast cancer, and in the one RCT that reported a significant mortality reduction for fecal occult blood screening in colorectal cancer, population differences led mortality comparisons to overestimate the effectiveness of screening. In lung cancer, no significant mortality reductions have been reported (to my knowledge), but in the two RCTs most directly addressing the effectiveness of chest radiograph (CXR) screening, population differences led mortality comparisons to underestimate the ability of CXRs to reduce the risk of dying of lung cancer. Although mortality is believed to be the best measure of outcome, not a single example can be cited as definitive proof of efficacy for any screening strategy. Thus, screening cannot be recommended for any cancer on the basis of consistent reductions in mortality in RCTs. Current policy, which calls for no early detection efforts for lung cancer, implicitly accepts the validity of two contradictory assertions. Conventional wisdom maintains that lung cancer is a highly virulent disease and that metastases are present at inception; accordingly, early detection is ineffective. However, RCTs suggest that lung cancer is an indolent disease and that radiographically detected lesions are clinically unimportant; accordingly, early detection is unnecessary. Such contradictions mandate some rethinking of the fundamental assumptions underlying screening evaluation. Considerable evidence suggests that annual CXR screening could result in a dramatic reduction in lung cancer mortality in our society. However, proper interpretation of the data depends completely on how screening effectiveness is measured. Given the enormous public health importance of this issue, a consensus conference is recommended to determine whether lung cancer screening can save lives.

Screening for Lung Cancer: Another Look; A Different View

There is widespread acceptance that screening for lung cancer is not indicated, to our knowledge, because no randomized trial has demonstrated a reduction in mortality as a result of screening. The objectives of this work are to review prospective studies on lung cancer screening and to analyze the extent to which known biases may have influenced observed results. Four randomized controlled trials have been conducted. The Memorial-Sloan Kettering and Johns Hopkins Lung Projects compared annual chest radiographs (CXRs) in a control group with CXRs and sputum cytologic findings in an experimental group. Although both studies failed to demonstrate any difference in outcome by the addition of cytologic study to CXR, long-term survival in both studies was approximately three times that predicted by other data. Accordingly, these results are at least consistent with the hypothesis that the screening CXRs may have improved survival. Two randomized trials, the Mayo Lung Project and the Czechoslovak study, compared regular and frequent rescreening CXRs in an experimental group with sporadic and/or infrequent rescreening in a control group. Both the Mayo and Czech studies demonstrated a striking advantage for screening with respect to stage distribution, resectability, survival, and fatality. Nevertheless, mortality was somewhat higher in the screened groups in both studies. Survival and fatality comparisons in randomized trials can be confounded by overdiagnosis bias, lead-time bias, and length bias, while mortality is not subject to these biases. Accordingly, it is believed that a mortality reduction represents the strongest evidence for screening efficacy. Mortality is directly proportional to cumulative incidence. In both the Mayo and Czech studies, incidence of lung cancer was higher in the screened group. The higher cumulative incidence (which in the Mayo Lung Project was statistically significant) made possible the discordant findings of superior survival/fatality and inferior mortality in the screened populations. Overdiagnosis has been widely accepted to account for the "missing cases" in the control populations in the Mayo and Czech studies. However, epidemiologic and autopsy evidence as well as an outcome analysis of unresected early-stage screen-detected lung cancer demonstrates that screening does not lead to the overdiagnosis of lung cancer. Similarly, lead-time bias and length bias cannot account for the outcome differences in the Mayo Lung Project or Czech study. If survival and fatality comparisons (which suggest a striking benefit from screening) are not biased, then mortality comparisons (which suggest no benefit) cannot accurately reflect lung cancer death rates in these trials. Population heterogeneity may provide an explanation for how outcome differences may have been misrepresented by mortality comparisons in these two trials, as well as other large population-based randomized studies. Periodic screening CXRs lead to clinically meaningful improvements in stage distribution, resectability, survival, and fatality in lung cancer. Mortality reductions have not been demonstrated, but mortality did not accurately reflect lung cancer death rates in the Mayo Lung Project and Czechoslovak study. Accordingly, reconsideration of the desirability of periodic CXR screening may be appropriate for individuals at high risk of lung cancer.

Chest X-ray Screening for Lung Cancer

Strauss and colleagues (CHEST 1995;[suppl]107:270S-79S) claim that periodic chest x-ray screening for lung cancer should be reconsidered because two randomized trials (their references 10-13) have shown improvements in stage distribution, resection rate, and survival. With convoluted reasoning, they dismiss overdiagnosis and lead-time bias as explanations for these results. They discard the well-accepted dictum that screening has had no impact on lung cancer population mortality rates, which are conclusions supported by the two studies they reviewed.

Screening for Lung Cancer Re-examined: A Reinterpretation of the Mayo Lung Project Randomized Trial on Lung Cancer Screening

In the 1970s, three randomized trials, each involving approximately 10,000 male subjects, were carried out to determine the value of lung cancer screening in cigarette smokers. These studies have been widely interpreted as providing strong evidence that screening does not contribute to decreased death rates, and, accordingly, the American Cancer Society firmly recommends against lung cancer screening. No randomized trial, however, has evaluated screening for lung cancer in women, who currently comprise 39% of the lung cancer population. The trials performed at Memorial-Sloan Kettering and at Johns Hopkins were designed so that subjects were randomized to either a single screen (annual chest x-ray alone) or a dual screen (annual chest x-ray and every-4-month sputum cytology) group. While the results of both revealed no difference in mortality between the groups, these studies were designed to examine the value of sputum cytology, and no conclusion regarding the efficacy of chest x-rays can be inferred. In the Mayo Lung Project, patients were randomized to a screened group in which a chest x-ray and sputum cytology were obtained every 4 months or to a control group in which an annual chest x-ray and cytology were simply recommended. The results indicate that after 6 years, more lung cancers were detected among the 4,618 men in the screened group (206 cases, 4.46%) than in the 4,593 men in the control group (160 cases, 3.48%). The screened group showed a superiority over the control subjects in resectability rate (46% vs 32%) and 5-year survival (33% vs 15%). However, the number of cancer deaths was slightly greater in the screened (122) than in the control group (115), and, consequently, the mortality rates were not significantly different among the groups. An "overdiagnosis bias" has been suggested to account for the increased number of lung cancers detected in the screened vs the control population in the Mayo Lung Project. This explanation is statistically plausible, but, given the status of lung cancer as the most lethal of malignancies, is biologically unlikely. An alternative hypothesis is that chance alone might have accounted for the observed 0.98% difference in lung cancer detection rates. Were this the case, then 46 additional cases would have been detected in the control group had this chance event not occurred.(ABSTRACT TRUNCATED AT 400 WORDS)

Survival for clinical Stage I lung cancer not surgically treated: Comparison between screen-detected and symptom-detected cases

To assess the extent of overdiagnosis bias in lung cancer screening, clinical Stage I lung cancer cases detected by chest radiograph examination, with histologic or cytologic evidence of malignancy and not treated by surgical operation, were followed up for more than 10 years. Of 1297 screen-detected and 1297 symptom-detected cases collected from 20 institutions, 42 screen-detected and 27 symptom-detected cases satisfied the study criteria. In about half of the cases, the patients had no contraindication for surgical treatment, but they refused surgical procedure. All such patients from the screen-detected and symptom-detected groups died within 122 and 67 months, respectively, of diagnosis. Among the screen-detected and symptom-detected cases, 80% and 81%, respectively, of the patients died of lung cancer. The median survival time was 25 and 13 months for those in the screen-detected and symptom-detected groups, respectively. The difference in survival was statistically significant between the two groups, which indicated the effect of lead time and length-biased sampling. Analysis of the causes of death other than lung cancer showed that there was no difference in the observed cumulative rates of deaths of other causes between the two groups, and these figures were almost the same as those expected from the general population. This indicates that overdiagnosis bias would be minimal in screen-detected lung cancer cases detected by chest radiograph examination.

Survival for clinical stage I lung cancer not surgically treated. Comparison between screen-detected and symptom-detected cases

To assess the extent of overdiagnosis bias in lung cancer screening, clinical Stage I lung cancer cases detected by chest radiograph examination, with histologic or cytologic evidence of malignancy and not treated by surgical operation, were followed up for more than 10 years. Of 1297 screen-detected and 1297 symptom-detected cases collected from 20 institutions, 42 screen-detected and 27 symptom-detected cases satisfied the study criteria. In about half of the cases, the patients had no contraindication for surgical treatment, but they refused surgical procedure. All such patients from the screen-detected and symptom-detected groups died within 122 and 67 months, respectively, of diagnosis. Among the screen-detected and symptom-detected cases, 80% and 81%, respectively, of the patients died of lung cancer. The median survival time was 25 and 13 months for those in the screen-detected and symptom-detected groups, respectively. The difference in survival was statistically significant between the two groups, which indicated the effect of lead time and length-biased sampling. Analysis of the causes of death other than lung cancer showed that there was no difference in the observed cumulative rates of deaths of other causes between the two groups, and these figures were almost the same as those expected from the general population. This indicates that overdiagnosis bias would be minimal in screen-detected lung cancer cases detected by chest radiograph examination.

Evaluating the evidence for the effectiveness of prevention

This report reviews a number of issues that should be considered in evaluating the evidence for the effectiveness of primary and secondary prevention. In evaluating primary prevention, changes in disease incidence are the ideal index but mortality may also be a useful surrogate in certain cases. Both relative risk and attributable risk are valuable concepts in evaluating the benefits of primary prevention. In evaluating screening, the most notable approach to secondary prevention, certain potential biases should be taken into account. These include selection biases due to referral or length-biased sampling, lead time bias, and overdiagnosis bias. An understanding of the methodologic issues involved in evaluating prevention will help ensure the validity of any conclusions drawn regarding the effectiveness of preventive measures and programs.