A feasibility study to assess the validity of administrative data sources and self-reported information of breast cancer survivors

Abstract

Background: Cancer survivorship has increasingly become the focus of research due to progress in early detection and advancements in the therapeutic approach, but high-quality information sources for outcomes, potential confounders and personal characteristics present a challenge. Few studies have collected breast cancer care data from mixed data sources and validated them, and to the best of our knowledge, none so far have been conducted in Israel, where National Health Insurance Law assures universal health care, delivered through four health care funds with computerized administrative, pharmaceutical and medical databases. This validation study is aimed to assess the accuracy and completeness of information on cancer care and health outcomes using several research tools, before embarking on a full-scale study aimed to evaluate the long-term treatment-related health adverse outcomes in a cohort of breast cancer survivors.

Methods: One hundred twenty randomly sampled female patients diagnosed with primary breast cancer in years 2000-2010 in northern Israel, who are members of the "Leumit" healthcare fund, were included. Data sources included "Leumit" medical records, the National Cancer Registry and a self-report questionnaire. The questionnaire was completed by 99 % of the women contacted. The accuracy of the information regarding cancer care was assessed with the reference standard set as one of the research tools, varying per the characteristic being under investigation. For example: health outcomes and medical history were validated against "Leumit" medical records, while construct validity of the self-reported questionnaire served to assess the prevalence of chronic pain. Agreement, predictive values, correlations, and internal consistency were calculated. Logistic regression models were constructed to assess potential predictors of correct responses.

Results: The overall level of agreement (Kappa) was almost perfect for demographics and outcomes, above 0.8 for treatments and chronic pain, while only fair to moderate for most of the self-reported medical history. Correct responses of medical history were associated with Jewish ethnicity, recency of breast cancer diagnosis, and family history of cardiovascular disease. The internal consistency of the quality-of-life scale was above 0.9.

Conclusion: In the absence of a national registry for cancer care, a mixed methodology for data collection is the most complete source.

Trial registration: Trial registration number Not available. This is an observational study with prospective data collection and no intervention; therefore, trial registration number is not required.

Keywords: Administrative data; Agreement; Breast cancer; Medical record; Self-report; Validity.

Fig. 1

Survey sample flow chart

Fig. 2

Distribution of treatments among 83 one-year breast cancer survivors. Notations: Rx, radiotherapy; Chem, chemotherapy; Hormone, hormone therapy. The proportions do not sum 100 % because more rare combinations of treatment (i.e., surgery only; etc.) are not presented

Fig. 3

Validity and completeness of patient continuous features across data sources: a, Mountain plot analysis of agreement between INCR and LHS on age at breast cancer diagnosis. Median = (−0.04y), percentiles: 2.5th = (−9.2y) 97.5th = 0.2y; b, Mountain plot analysis between INCR and questionnaire on breast cancer diagnosis year. Median = 0y, percentiles: 2.5th = (−1y) 97.5th = 1y; c, Bland Altman analysis of agreement between LHS and QUES on body mass index (BMI). The mean BMI difference = 0.5 kg/m² (mean weight and height difference, 0.9 kg and 0.6 cm, respectively), and the 95 % limits of agreement from −2.3 kg/m² to 3.3 kg/m² (Measured BMI was up to one year from self-reported BMI)