2.1. Study design and patient selection

Analysis was conducted on a pseudonymised, retrospective database containing peripheral venous blood sample results between 1 January 2004 and 1 September 2013 from 21,020 patients, all of whom had received chemotherapy treatment at the Leeds Cancer Centre, Leeds Teaching Hospitals Trust (LTHT). The LTHT results server receives blood test results from the pathology laboratories and displays them in the electronic patient recording system (Patient Pathway Manager (PPM)) [16], [17]. A pseudonymised extract was taken and inserted into a research database. No identifiable data was contained within the dataset and the research was sanctioned under the information governance procedures of LTHT, with data extraction pseudonymisation procedures as agreed with the Caldicott Guardian and with formal approval from a national research ethics committee (NHS Grampian ID: 13/NS/0128). No patients were excluded based on their chemotherapy treatment, demographic information, diagnosis or timing of treatment.

Blood counts were measured from EDTA venous whole blood samples obtained for the purposes of routine clinical care, and taken at any time in relation to chemotherapy delivery. All samples were submitted for a full blood count analysis, including a five-part differential on a Siemens ADVIA 120 analyser (Siemens Healthcare Diagnostics, Erlangen, Germany) until August 2004 and subsequently on the Siemens ADVIA 2120 analyser; both instruments employ the same method principles. All instruments were subjected to multiple quality control (QC) checks each day according to standard laboratory protocols, and the laboratory participated in the United Kingdom National External Quality Assessment Service (UKNEQAS) external quality assurance scheme.

Data of interest included the eosinophil, basophil and neutrophil counts, with the sum of these three parameters being taken as the granulocyte count (calculated in Microsoft SQL Server). Lymphocyte and monocyte results were also extracted for analysis. As within-day timing information was not available, if a patient had more than one blood test on a given day all data for that day was excluded to avoid ambiguity as to which result should be taken as the true value for that day.