On 2014 Nov 17, Peter H. Asdahl commented:

The study by Villani et al. addresses the issue of cancer surveillance among TP53 mutation carriers with Li-Fraumeni syndrome. The authors interpreted data from an institutionally-devised cancer surveillance program. Villani et al. concluded that their surveillance approach is feasible, and may lead to a reduction in mortality. In addition, the authors emphasized the importance of genetic testing for early detection. Nonetheless, certain results of the study warrant further discussion.

Villani et al. used an observational design to address their aim, where mutation carriers were given the option of participating in the surveillance or non-surveillance (i.e. standard clinical practice) groups, and were subsequently followed for incident cancers. Given the goal of systematic cancer surveillance, we would expect a higher yield of incident cancers in the surveillance group. Nevertheless, the result from the Villani et al. study contradicts expectation. Specifically, one or more tumors were detected among seven of the 18 (39%) participants who opted for surveillance in contrast to 10 of the 16 participants (63%) who opted for non-surveillance. Given insufficient data in the published report by Villani et al. to estimate person-time contribution for individuals in each group, we assume equivalent follow-up between groups for the purpose of our illustration. Consequently, a patient-level comparison of the two groups yields a 24% reduction in the absolute risk of incident cancers in the surveillance group (risk difference= -24%, 95% confidence interval: -56% to 9%). We would not expect a negative risk difference in this scenario, which raises concerns about potential biases such as healthy user bias or confounding by functional status [1].

Healthy user bias and confounding by functional status are often overlooked but legitimate problems in observational studies of screening effects. For example, if participants in the non-surveillance group had more functional impairments (e.g. related to underlying but undetected cancer) which resulted in opting for standard clinical care rather than systematic surveillance, then eventual detection of underlying cancer in this group may result in a higher yield compared with the surveillance group. Some evidence for this phenomenon may be apparent by the lower than expected survival among participants with incident cancers in the non-surveillance group. In particular, three-year survival among the non-surveillance group was 21%, which is low even for aggressive tumors.

An ideal screening program aims for early detection of cancer with subsequent reduction in cancer-related mortality. Although we do not reject the notion that the screening program in this study detected asymptomatic tumors, the interpretation that the authors’ surveillance program is superior to non-surveillance is not supported by the results. To investigate a possible effect of screening in this population, a more robust study design with random allocation to intervention may be less sensitive to bias.

REFERENCE 1. Shrank WH, Patrick AR, Brookhart MA. Healthy user and related biases in observational studies of preventive interventions: a primer for physicians. J Gen Intern Med. 2011 May;26(5):546-50.

On 2014 Nov 17, Peter H. Asdahl commented:

The study by Villani et al. addresses the issue of cancer surveillance among TP53 mutation carriers with Li-Fraumeni syndrome. The authors interpreted data from an institutionally-devised cancer surveillance program. Villani et al. concluded that their surveillance approach is feasible, and may lead to a reduction in mortality. In addition, the authors emphasized the importance of genetic testing for early detection. Nonetheless, certain results of the study warrant further discussion.

Villani et al. used an observational design to address their aim, where mutation carriers were given the option of participating in the surveillance or non-surveillance (i.e. standard clinical practice) groups, and were subsequently followed for incident cancers. Given the goal of systematic cancer surveillance, we would expect a higher yield of incident cancers in the surveillance group. Nevertheless, the result from the Villani et al. study contradicts expectation. Specifically, one or more tumors were detected among seven of the 18 (39%) participants who opted for surveillance in contrast to 10 of the 16 participants (63%) who opted for non-surveillance. Given insufficient data in the published report by Villani et al. to estimate person-time contribution for individuals in each group, we assume equivalent follow-up between groups for the purpose of our illustration. Consequently, a patient-level comparison of the two groups yields a 24% reduction in the absolute risk of incident cancers in the surveillance group (risk difference= -24%, 95% confidence interval: -56% to 9%). We would not expect a negative risk difference in this scenario, which raises concerns about potential biases such as healthy user bias or confounding by functional status [1].

Healthy user bias and confounding by functional status are often overlooked but legitimate problems in observational studies of screening effects. For example, if participants in the non-surveillance group had more functional impairments (e.g. related to underlying but undetected cancer) which resulted in opting for standard clinical care rather than systematic surveillance, then eventual detection of underlying cancer in this group may result in a higher yield compared with the surveillance group. Some evidence for this phenomenon may be apparent by the lower than expected survival among participants with incident cancers in the non-surveillance group. In particular, three-year survival among the non-surveillance group was 21%, which is low even for aggressive tumors.

An ideal screening program aims for early detection of cancer with subsequent reduction in cancer-related mortality. Although we do not reject the notion that the screening program in this study detected asymptomatic tumors, the interpretation that the authors’ surveillance program is superior to non-surveillance is not supported by the results. To investigate a possible effect of screening in this population, a more robust study design with random allocation to intervention may be less sensitive to bias.

REFERENCE 1. Shrank WH, Patrick AR, Brookhart MA. Healthy user and related biases in observational studies of preventive interventions: a primer for physicians. J Gen Intern Med. 2011 May;26(5):546-50.