Omitting missing data non consented patients had features as
Omitting missing data, non-consented patients had features asso-ciated with a poorer prognosis (Table 3); they were more often clini-cally rather than screen detected, and 13% had metastatic disease compared with 3% of consented patients. Non-consented patients more often had no primary surgery (26.2% compared with 4.0% of consented patients), and less frequently had chemotherapy, radiotherapy, hor-monal therapy, or biological therapy (all p < 0.001), in part because more non-consented patients declined these treatments. They had more often positive tumour margins (p = 0.05). However, they did not differ by tumour size, histological type, or hormone receptor status, and
tended to have fewer nodes positive. Higher proportions of non-con-sented patients had missing data on most clinical variables (Table 4). The results of the sensitivity analysis of survival from 90 days, to allow for delayed consent completion up to this time, were very similar.
There were also 1642 women with ductal carcinoma in situ (DCIS) disease, of whom 1397 (85.1%) were consented and 245 (14.9%) were not. Mean follow-up was 5.7 years, median 5.2 years and maximum 13.7 years. Survival overall was as expected very high, 96.1% at 5 years for all patients. As with invasive disease, survival for DCIS at each time point and overall was lower in non-consented than in consented pa-tients (Table 1), logrank = 9.8, p = 0.002. Survival rates for consented patients were higher than for all patients at all time points, but the differences were less than 1%.
Cancer registries have been very useful for assessing cancer trends, management, and outcomes. Large national or regional registries may have legislative authority to operate without requiring each patient’s consent to be included. However, for clinically based registries, consent of individual patients may be required for data to be stored and to be used for research purposes. Patients who consent may be a selected altruistic group and if so, could give rise to selection biases in analyses. This study uses an opportunity to compare in detail patients who gave informed consent against those who did not.
The clinical registry in this study is probably typical of a registry designed by interested clinicians to bring together the data on patients’ demographic, clinical, pathological and treatment details that are al-ready recorded in various hospital records. Outcome data was expanded by regular UNC1999 to the national cancer registry and mortality data done securely by the Ministry of Health using each patient’s unique New Zealand national health indicator identifier. Follow-up for mor-tality is complete unless the patient emigrates and dies outside New Zealand; ‘emigration’ estimates for this age group are 0.9% per year, but these are overestimates as returns after 16 months’ absence are not recorded . The registry had very limited funding. The requirement for informed consent for inclusion on the registry meant that patients were approached by clinical staff usually at their first visit to a
Clinical features and treatment of consented and non-consented patients.
Consented, Non- p-value
Resection margin No surgery 0.048
those who had
Subjects with missing data excluded (see Table 4 for details).
specialist clinic. For some patients the process was delayed but the consent process was completed by first post-surgery therapy, within 90 days of diagnosis. Short delays may have been influenced by the se-verity of disease, but the analysis of survival from 90 days after diag-nosis showed no substantial differences to the main analysis. Patients who did not give consent may have declined it, or may not have been offered the relevant information and consent forms; the data does not distinguish between these.
The results here support other results showing that requiring con-sent results in only partial participation and this is differential: subjects with more severe disease and worse outcomes tend to be excluded. In this study there was 10.4% non-consent for invasive cancer, resulting in a 2.4% upward bias in 5-year survival if this is based only on consenting patients. Small variations in survival are important: for example, a 2.4% difference is similar to the improvement in breast cancer survival seen Cancer Epidemiology 58 (2019) 178–183
The size of the bias depends on the proportion of patients excluded by non-consent and by how this selection influences this group, which will vary with each registry. Thus, in an observational study of breast cancer in the UK, embryo sac was demonstrated that the use of the dataset re-quiring consent omitted information about many women with locally advanced or metastatic cancer, and also underrepresented women in deprived social groups . A clinical registry of acute renal disease in the US requiring no interventions but requiring informed consent only achieved 52% completeness . A major multicentre effort to record information of the natural history and treatment of stroke in Canada failed because of the requirement for informed consent, so that initially only 39% of eligible patients, and after simplification of the require-ments only 50% of eligible patients, give informed consent for the data to be used; and the costs involved in obtaining consent formed a high proportion of the total costs of the endeavour. The authors argued that the avoidance of informed consent from minimal risk observational research is essential .