An analysis of survival differences between clinically and screen‐detected cancer patients

Abstract

Survival differences between clinically and screen-detected cancer patients partly result from biases. Well known are lead-time, length bias and overdiagnosis. The survival of the clinically detected patients in the study group of the HIP breast cancer screening project is corrected for these biases. The resulting survival curve is only slightly worse than the survival of the screen-detected patients. This suggests a very modest mortality reduction by screening. A much larger reduction is obtained from an analysis of the complete HIP results, including those of the control group. It is concluded that a large unexpected selection bias is present. This bias would not have been detected if the HIP study had not contained a randomized control group. A misleading and pessimistic conclusion on the effectiveness of breast cancer screening would thus have resulted. This conclusion reinforces the need for randomized studies.

Instrumental variables when evaluating screening trials: estimating the benefit of detecting cancer by screening

“…De"ne G as in (6) for any t*'0 de"ne M G as in (5). Then (7) states the ignorability condition for the cancer cohort, and (8) states that the cancer statuses at t* must not depend on the treatment assignment at the time the cancer cohort is constructed. Expression (9) says subjects who are screen detected do not have di!erent outcomes that those who are not screen detected.…”

Estimating life expectancy and related probabilities in screen‐detected breast cancer patients with restricted follow‐up information

“…The ÿrst mass screening programme for breast cancer was the so-called HIP study (Health Insurance Plan of Greater New York) in the early 1960s [1]. Several analyses have been made on the HIP survival data (Walter and Stitt [12], Habbema et al [13], Aron and Prorok [14], Kafadar and Prorok [15] and Xu and Prorok [16]). All these papers present methods to estimate the breast cancer mortality reduction through screening, by comparing the survival of screen-detected cancer cases to the survival of interval cancer cases, non-participant cancer cases and cancer cases in randomized control groups.…”

“…The mean duration of the lead time was estimated as 1.71 years. The hazard rate of dying from breast cancer in the post-lead time was estimated at = 0:03664 [13].…”

“…When the survival of breast cancer cases in the control group was used as a reference, 25.4 breast cancer deaths were prevented, which corresponds with a reduction of 47 per cent. Habbema et al [13] reported ÿve prevented deaths through screening when the survival of clinical cases in the study group was used as a reference, versus 40 prevented deaths when the survival of cases in the control group was used. The large di erences in the prevented number of deaths between the calculations by Walter and Stitt and Habbema et al can be explained by di erent assumptions.…”

“…The ideal way to assess breast cancer mortality reduction by screening is a randomized trial, in which mortality in the group o ered screening is compared to the mortality in the group not o ered screening. Some authors used survival after diagnosis from several types of cancer to estimate mortality reduction by screening [12][13][14][15][16]. They mentioned various problems, such as lead time, length time bias, selection bias and overdiagnosis, which complicate interpretation of the estimated mortality reduction.…”