Selection bias may affect breast cancer screening effectiveness

A study out of Denmark found that a breast cancer screening program's purported effectiveness could be strongly influenced by the patients selected for screening. The researchers suggest that program detection rates should be interpreted with caution as they may be biased by eligibility criteria.

Denmark is one of the countries with the highest breast cancer incidence rates in the world. Mammography screening was introduced in the municipality of Copenhagen in 1991 and the county of Fyn in 1993. Nationwide rollout took place in 2007 through 2010, but the screening programs have organizational differences regarding the definition of eligible women and invitation procedures.

A team led by Laia Domingo, predoctoral researcher at the Department of Epidemiology and Evaluation, Hospital del Mar-IMIM, Barcelona and the Department of Public Health, University of Copenhagen, sought to determine the rates of screen-detected, interval cancers and cancers diagnosed outside screening over 17 years in Denmark, emphasizing the influence of organizational differences in the screening programs to changes of breast cancer detection patterns in the population (Acta Oncologica, January 2013, Vol. 52:1, pp. 48-56).

The researchers found the start of screening resulted in a peak of cancer prevalence in participants, followed by a decline to a fairly stable detection rate in subsequent invitation rounds. A similar pattern was found for breast cancer incidence in nonscreened women. In Fyn, nonscreened women had a higher cancer prevalence rate than screening participants during the first three invitation rounds, according to the researchers.

"As expected due to the lead time, the start of organized screening in Denmark resulted in a prevalence peak of screen-detected breast cancers," the researchers wrote. "It was highly surprising, however, that the start of screening resulted also in a 'prevalence peak' among nonscreened women."

The interval cancer rate was lower in Copenhagen than in Fyn, with an increase over time in Copenhagen but not in Fyn. Screen-detected cancers showed tumor features related with a better prognosis than tumors detected otherwise, as more than 80% were smaller than 20 mm and were estrogen receptor-positive, they added.

Screen-detected tumors in Copenhagen tended to be smaller and less invasive than those detected in Fyn, as 41% in Copenhagen versus 35% in Fyn were 10 mm or smaller, and 15% versus 12%, respectively, were in situ carcinomas, according to the researchers.

Strengths and weaknesses

The long study period and the completeness and accuracy of the data were the primary strengths of the study. The current work was based on data from organized population-based screening and data from a population-based cancer register, the researchers added.

"This allowed accurate identification of breast cancer cases detected both inside and outside of screening, and the use of unique personal identification numbers ensured complete linkage," Domingo and colleagues wrote. "All screening data were obtained from computerized records and came from two screening programs in Denmark whose performance indicators reached the European standards."

The study's limitations included the inability to divide interval cancers into those coming from false-negative screens and those developed after screening; misclassification of the detection mode could not be excluded.

"Some interval cancers could be classified as screen-detected if symptomatic women waited for the screening visit instead of making an immediate appointment with a physician," the researchers wrote. "This misclassification would attenuate differences on tumor characteristics between screen-detected and interval cancers."

Also, at the beginning of the Copenhagen program, only women treated for breast cancer within the past 18 months were considered noneligible, and 95% of the target population was invited, with 71% of the invited women participating. The "prevalence peak" in nonscreened women was therefore unlikely to be explained by a selection bias, they added.

Increased awareness of breast cancer along with the start of the screening program could explain the prevalence peak.

"Women with symptoms might then have sought help before they were invited to the program," Domingo and colleagues wrote.

The situation at the start of the screening program in Fyn was different. Only 12% of the targeted women were identified as noneligible. In addition to women already diagnosed with breast cancer within the last five years, the noneligible women from 1997 onward also included those with a prior excision of in situ or nonspecified neoplasms, and women with a recent mammogram could cancel participation.

"In spite of a high breast cancer incidence in Danish women, screening has allowed detection of breast cancers at early stages," the authors concluded. "Nonetheless, the proportion of breast cancers detected outside screening remained high, and these cases had less favorable prognostic characteristics than the breast cancer inside and outside screening in Denmark screen-detected cases."

High data quality is needed to evaluate mammography screening, the author continued. They recommend that all data should be stored with personal identifiers, and all files, including the administrative files on targeted women and invitations, should be kept.

"Our data illustrated that organizational aspects and screening setting of the programs, as invitation criteria and participation rates, strongly influenced the detection rate which is used as the standard indicator for the outcome of screening," they wrote. "It is therefore pertinent that evaluation of screening starts from the target population, and that the eventual selection in participating women is taken into account in the interpretation of the screening outcome."

Page 1 of 1243
Next Page