The protocol for this systematic review is available online and as an appendix (see Appendix A).
3.3.1 Scoping search
Before the full search strategy for this systematic review was finalised, a scoping search was carried out. This included informal, unstructured searching using internet search engines and Pubmed, and key terms relating to thrombocytosis, cancer, and primary care. A few key papers were known prior to beginning the review. The scoping search tested the search strategy to ensure it was effective at identifying the known key papers. If they were found by the search strategy, it is possible to have more confidence in the search strategy’s ability to identify unknown yet relevant papers.
3. Systematic review of thrombocytosis and cancer diagnosis
3.3.2 Search strategy
The search strategy was developed using three key terms: “thrombocytosis”, “cancer”, and “primary care”. A range of additional search terms were derived from each of these; for thrombocytosis the following terms were used: “thrombocytosis” or “platelet” or “thrombocyte” or “thrombocyte count”. For cancer, “neoplasm” was also included. Additional terms for primary care included “primary medical care” or “family practice” or “family medicine”. The full search strategy is included in Appendix A.
The search was limited to English language papers due to a lack of available trans- lating facilities. The search was also limited to results published in the last 30 years (the scoping search found no results beyond 30 years). The following databases were searched: EMBASE (OvidSP); Medline (Ovid); Web of Science, The Cochrane Library. Forwards and backwards citation searching was carried out on included papers. This involved checking the papers referenced in included studies for eligibility for inclusion, and examining later published works for eligibility which had referenced the included papers. In addition to searching electronic databases, a range of national and interna- tional experts were contacted to ask if they knew of any other relevant studies. The need for this strategy became apparent as, prior to starting the systematic review, I was aware of four studies that had collected data relevant to the systematic review but had not published the results because they were negative, and one further study which had collected but not analysed platelet count data. Publication bias is a well- documented limitation of systematic reviews, and it was a strength of this review that experts were contacted to ask about what data they may have that were relevant to the review, even if the results were negative. In addition to approaching other researchers directly, I attended a number of national and international conferences throughout the term of this PhD, delivered several oral and poster presentations, and had numerous discussions with other researchers at these events. Through this process I was able to determine if anyone else within the cancer diagnostics field was pursuing this line of investigation, or had collected relevant data.
All of the relevant papers identified by the literature search were exported to End- note X5 and de-duplicated. Relevant papers identified through non-literature search were included.
3.3.3 Study selection
The search aimed to identify any study that had investigated the association between thrombocytosis and a new diagnosis of cancer of any type in a primary care setting.
3. Systematic review of thrombocytosis and cancer diagnosis
Adults aged ≥ 40 years; Primary care setting;
Observational, case-control or cohort study, or any literature review. Exclusion criteria were:
Adults under 40 years of age. These were excluded because cancers in this age group tend to be familial or atypical.
Studies that had investigated platelet count as a prognostic tool or guide for cancer therapy. These were excluded as the systematic review specifically aimed to find studies that had used thrombocytosis in a diagnostic capacity.
Titles and abstracts were screened by myself and ES, a supervisor of this PhD and co-author on the subsequent publication, and full text articles were retrieved and assessed for inclusion.
3.3.4 Study quality assessment
The quality of included studies was assessed using the Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) tool (Whiting et al., 2011). This is a tool for assessing the quality of diagnostic test studies. Although the overall aim of this PhD is to investigate thrombocytosis as a marker of cancer, not as a diagnostic test for cancer, this quality assessment tool was chosen as it best fits the type of studies that the review expected to identify and the types of studies that were found in the scoping search. The QUADAS-2 tool assesses studies for risk of bias and applicability concerns. The domains within these two areas are:
Risk of bias:
Patient selection - could the patient selection procedure have introduced bias? Index test - could the conduct or interpretation of the index test have introduced
bias?
Reference standard - could the reference standard, its conduct, or interpretation have introduced bias?
Flow and timing - could the patient flow have introduced bias? Are there any concerns about the time interval between the index test and the reference stan- dard? Are all patients receiving the same reference standard and are all patients included in the analysis?
3. Systematic review of thrombocytosis and cancer diagnosis
Applicability concerns:
Patient selection - do the included patients and the setting match the review question?
Index test - are there any concerns that the index test differs from that in the review question?
Reference standard - are there any concerns that the target condition, as defined by the reference standard, does not match the review question?
3.3.5 Data extraction
A range of data relating to the number of patients with and without the cancers of interest, and the number of these with and without thrombocytosis prior to diagnosis were extracted from papers or from personal communications with the authors onto custom-made data extraction forms (see Appendix B). These forms also held details of the study design, patient or data sources, and patient characteristics. Data extraction was primarily carried out by myself, and was checked by WH, a supervisor of this PhD and co-author on the publication which resulted from this systematic review.
3.3.6 Data analysis: narrative synthesis
Two methods were used to collate and analyse data. Firstly, the extracted data were drawn together in a narrative synthesis following the general framework set out in the Centre for Reviews and Dissemination’s Guidance for Undertaking Reviews in Health- care (Centre for Reviews and Dissemination, 2009). In the first instance, a preliminary synthesis of findings was carried out which involved a textual description of studies, tabulating raw data and key features of studies.
The raw data extracted from each study were used to calculate the likelihood ratio (LR: the probability of raised platelets in patients with cancer divided by the probability of raised platelets in patients without the disease) and the positive predictive value (PPV) using Bayes’ Theorem. The PPV reflects the probability that a patient with a positive test result has the disease in question. It is calculated as the proportion of patients with a positive test result who have the disease in question. This approach to calculating PPVs is used with case-control study data as it accommodates the lower prevalence in the general population, a necessary adjustment as prevalence is artificially high in case-control studies. Thus, PPVs were calculated as the LR multiplied by the prior odds of the disease (Knottnerus, 2002); in this case incidence data were used to calculate prior odds.
3. Systematic review of thrombocytosis and cancer diagnosis
The analysis examined two subgroups which were defined based on (1) whether the cancer being studied affects both sexes or only one (this latter group transpired to be females only as no male-only cancer sites were identified by the searches), and (2) the data sources used by the studies. The relationships within and between studies were explored, and the strength and robustness of the evidence was assessed.
3.3.7 Data analysis: meta-analysis
Where available, the odds ratios from relevant studies were pooled using the random ef- fects meta-analysis model in RevMan 5.3. (RevMan, 2014). The primary meta-analysis included all studies, and two sensitivity analyses examined the effect of pooling results from the two subgroups defined above. Heterogeneity in these models was examined us- ing I-squared (I2) statistic. The primary meta-analysis included several studies which had each investigated a specific cancer site. There were no studies that had investigated all or any type of cancer (two studies that have investigated the risk of any type of cancer in patients (Hippisley-Cox & Coupland, 2013a,b) did not include platelet count data) It is possible that the strength of any association between thrombocytosis and cancer may vary between different cancer sites. Furthermore, not all cancer sites had been investigated by studies identified by this review. 95% prediction intervals were calculated for pooled odds. There may not be one true single effect for the relationship between thrombocytosis and cancer; it may vary greatly between different types of cancer giving multiple different true effects. The 95% prediction interval quantifies the range of odds ratios across different cancer sites to provide an estimate of the interval in which the measure of association for another, as yet unreported, cancer site may lie. Rather than simply presenting the mean of the different effect of different cancer sites, the prediction interval reflects the range in which most of the different true effects lie.