Archives

  • 2019-07
  • 2019-08
  • 2019-09
  • 2019-10
  • 2019-11
  • 2020-03
  • 2020-07
  • 2020-08
  • Table br Baseline characteristics of women with stage I III

    2020-08-30

     Table 1
    Baseline characteristics of women with stage I-III breast cancer and breast cancer-free control women aged 64 years and below at diagnosis in the Breast Cancer Data Base Sweden.
    Characteristic Women with breast cancer Breast cancer-free women
    Age at diagnosis (years)
    Education
    Prior sick leave
    Prior hospitalization for one of the studied causes
    Pathologic TNM stage
    Tumor size (mm)
    Abbreviations: ALND, axillary lymph node dissection; ER, Concanamycin A receptor; IQR, interquartile range; SNB, sentinel node biopsy. a Index year for breast cancer-free women.
    diagnoses (Supplementary Table 1). The completeness of medical diagnoses was high; less than three percent of all records of granted disability pension lacked information on the main underlying diagnosis. The analysis of cause-specific sick leave was restricted to women diagnosed 2005 or later from when information on the diagnosis was available.
    2.3. Cause-specific death
    Information on cause of death was obtained from the Cause of
    Death Register and was categorized into death from cancer, car-diovascular disease, inflammatory disease, or other causes.
    2.4. Tumor and treatment characteristics
    From the quality registers, we extracted information on tumor size, estrogen-receptor (ER) status, lymph node involvement, type of surgery, type of axillary surgery, radiotherapy, chemotherapy, and endocrine treatment.
    2.5. Post-diagnostic intermediate events
    Information on hospital care was obtained from the Patient Register, which has nationwide coverage of inpatient hospitaliza-tions from 1987 and outpatient specialized care from 2001. We retrieved information on the first inpatient or outpatient hospital visit following the breast cancer diagnosis for the same set of conditions used for sick leave and disability pension, with the addition of metastatic disease and infectious diseases (Supplementary Table 1). Inpatient hospital visits before diagnosis were also Concanamycin A recorded to enable adjustment for previous medical history.
    2.6. Statistical analysis
    We performed two separate analyses. In the first analysis, which was based on data from women diagnosed between 2005 and 2012 (Supplementary Table 2), we estimated probabilities, duration, and hazard ratios (HR) of sick leave using a multi-state Markov model including disability pension and death. The multi-state model had one working state, eight sick leave states (one state for each cause), eight disability pension states, and four death states (Supplemental Fig. 1). All women started in the working state. Using the non-parametric Aalen-Johansen estimator [18], cause-specific plots of the transition probabilities were generated. Due to few events, the multistate model was simplified by combining causes of sick leave, disability pension, and death, reducing the number of transitions (Supplemental Fig. 2). By fitting flexible parametric survival models to each transition [19], we predicted model-based probabilities of being in a state, as well as length of stay in the sick leave states for women with and without breast cancer.
    In the second analysis, based on data on women diagnosed between 2000 and 2012 (Supplementary Table 3), we estimated cause-specific HR of disability pension associated with breast cancer, breast cancer treatment, and tumor characteristics using standard flexible parametric survival analysis, while censoring for death and the other causes of disability pension. We further examined the contribution of post-diagnostic intermediate events on disability pension attributed to cancer in women with breast cancer by including the first time point of the intermediate event as a time-varying covariate.
    In both analyses, the baseline hazard was modelled with restricted cubic splines with 3 or 5 degrees of freedom (df) using time since diagnosis as the underlying time scale. Both proportional and non-proportional models were fitted, with 1 or 3 df for the time-varying effect of exposures. The estimated HR express the rate of sick leave or disability pension receipt in women with breast cancer compared with breast cancer-free women, or by treatment or tumor characteristics. The following covariates were adjusted for: age at diagnosis (categorized in five-year intervals or contin-uous time using splines), calendar year of diagnosis (2000e2003, 2004e2007, 2008e2012), highest level of education, region of residence, sick leave in the period 366e730 days prior to diagnosis and hospitalization for the medical condition of interest in the five years prior to diagnosis. In models estimating HR by treatment and 
    tumor characteristics, tumor size, ER status, lymph node involve-ment, and treatment modality were also adjusted for. Models included either only tumor characteristics or both tumor charac-teristic and treatment modality.