Clergy

The study questionnaire asked about demographic variables, and factors known or suspected to affect the risk of breast cancer, including reproductive and menstrual history, exogenous hormone intake, exercise, benign breast disease, alcohol, smoking, some dietary variables, chest exposure to ionising radiation, variables related to the woman's own birth, childhood growth and puberty, height, weight, melatonin-related variables (e.g., shift work, exposure to light at night), occupation, socioeconomic variables, family history of cancer, and chronic diseases. We included questions on childhood and adolescent exposures and behaviours, as well as those in adulthood, and where appropriate (e.g., alcohol consumption, exercise) asked about these variables by age, to allow exposure histories to be built. For the same reason, information on selected exposures is being updated in follow-up questionnaires. The questionnaire also asked the women to self-measure, or ask others to measure for them, certain anthropometric variables – weight, waist circumference, hip circumference and arm span.
Information was sought within the questionnaire, and the participant's informed consent was sought, to enable medical and other relevant records to be located and examined for validation of certain data, and to obtain exposure details that were not possible to obtain from the questionnaires. For instance, information and consent were collected to allow examination of the woman's own birth records, mammograms and treatments. The questionnaire was designed for the subjects’ responses to be read by Optical Character Reading (OCR) software via a scanner (Readsoft, 2011 ), and, where possible, it asked the respondent to give exact replies to numerical questions, in order to maximise data detail, rather than offering multiple-choice boxes to select between pre-formed ranges. The software provides quality control of the OCR by two methods. First, any characters that the software has uncertainty in reading, or is unable to read, are flagged up to a clerical operative to read and enter the correct character. About 8% of completed fields need clerical intervention. Secondly, for alphabetic and numeric characters, but not for ticks, the software shows the operator all of the responses the subject has made for that character (e.g., all the 7s, all the 8s, etc.), for the operator to inspect and decide whether they are all consistent and have been read correctly. In addition, for certain key variables such as date of birth, the software was programmed to compel the operator to read and check the response for each study subject. We also programmed range and validity checks that show the operator for correction, invalid or unlikely values of characters or combinations of characters (e.g., values for ‘month’ greater than 12).

Data were extracted from two large healthcare databases, one maintained by CHKS Ltd and the other maintained by the Intensive Care National Audit & Research Centre (ICNARC). CHKS provides comparative benchmarking services to NHS trusts. Data are created by a clerical coding method, similar to Hospital Episodes Statistics. Validation is performed locally by the Trust and centrally by CHKS Ltd to provide a quality-assured dataset that can be used to inform managerial and clinical decisions. The ICNARC case mix programme collects data on consecutive admissions to participating adult, general intensive care units (ICUs) in England, Wales and Northern Ireland. Data are collated locally by trained dedicated staff and are subject to local and central internal error checks [15 (link)].
Data were extracted on all adult surgical admissions to hospital (CHKS data) and to the ICU (ICNARC data) for 94 NHS hospitals in England, Wales and Northern Ireland between January 1999 and October 2004 inclusive. These hospitals were selected because they contributed to both databases throughout the study period. Admissions involving endoscopy, day-case surgery, cardiothoracic surgery, neurosurgery, organ transplantation, obstetrics or the surgical management of burns were excluded. For brevity, procedures that satisfied the inclusion criteria are described as general surgical procedures.
There are 6,920 surgical procedure codes in the Office of Population Censuses and Surveys (now part of Office for National Statistics and Surveys) classification. Surgical admissions to hospital were identified in the CHKS database by the presence of one of 4,910 codes that satisfied the inclusion criteria. Where more than one surgical procedure was performed during the same hospital admission, only the first procedure was included in the analysis. Several alternative Office of Population Censuses and Surveys codes may exist for any given procedure. In order to reduce bias arising from discrepancies in the coding process, procedures were categorised into one of 372 Healthcare Resource Groups (HRGs) based on clinical similarity and resource homogeneity. Many Office of Population Censuses and Surveys codes and HRG codes specify the presence of a complicating medical condition, the complexity of surgery or a particular age group. HRGs were then ranked according to mortality rates. High-risk surgical procedures were prospectively defined as those procedures included in an HRG with a mortality rate of 5% or more. The remaining procedures were classified as standard risk.
Surgical admissions to the ICU were identified in the ICNARC database by the source of admission (either operating theatre or operating theatre via ward), and were only included if the primary reason for admission was not an excluded surgical procedure. ICU admissions were prospectively divided into admissions directly to the ICU following surgery and admissions to the ICU following a period of postoperative care on a standard ward. Where patients were readmitted to the ICU, only the first admission was included in the analysis.
Data are presented as the median (interquartile range). Categorical data were tested with the chi-squared approximation, and continuous data were tested with the Mann–Whitney U test. Analysis was performed using GraphPad Prism version 4.0 (GraphPad Software, San Diego, CA, USA). Significance was set at P < 0.05.

Once informed consent was obtained, paper-based questionnaires (MBI–HSS and PHQ-9) and a data collection form were self-reported and deposited into sealed, secure collection boxes that were collected by the primary investigator. Data were then recorded in an Excel spreadsheet. The analysed variables included sex, age, ethnicity, relationship status, job position, discipline, number of hours of overtime per week, number of hours remunerated work outside the public service (RWOPS) per week, academic and/or study time per week, hours per week spent on clerical work, psychiatric diagnosis, family history of psychiatric disorder, psychotherapy attendance, psychotropic medication and/or substance use in the past month and lifetime medication and/or substance use. An instrument was not scored if three or more items were missing; therefore, 327 MBI questionnaires and 335 PHQ-9 surveys were scored. Categories with n < 15 were not analysed.