Sarah J Price

Is omission of free text records a possible source of data loss and bias in Clinical Practice Research Datalink studies? A case–control study

Objectives To estimate data loss and bias in studies of Clinical Practice Research Datalink (CPRD) data that restrict analyses to Read codes, omitting anything recorded as text. Design Matched case–control study. Setting Patients contributing data to the CPRD. Participants 4915 bladder and 3635 pancreatic, cancer cases diagnosed between 1 January 2000 and 31 December 2009, matched on age, sex and general practitioner practice to up to 5 controls (bladder: n=21 718; pancreas: n=16 459). The analysis period was the year before cancer diagnosis. Primary and secondary outcome measures Frequency of haematuria, jaundice and abdominal pain, grouped by recording style: Read code or text-only (ie, hidden text). The association between recording style and case–control status (χ2 test). For each feature, the odds ratio (OR; conditional logistic regression) and positive predictive value (PPV; Bayes’ theorem) for cancer, before and after addition of hidden text records. Results Of the 20 958 total records of the features, 7951 (38%) were recorded in hidden text. Hidden text recording was more strongly associated with controls than with cases for haematuria (140/336=42% vs 556/3147=18%) in bladder cancer (χ2 test, p<0.001), and for jaundice (21/31=67% vs 463/1565=30%, p<0.0001) and abdominal pain (323/1126=29% vs 397/1789=22%, p<0.001) in pancreatic cancer. Adding hidden text records corrected PPVs of haematuria for bladder cancer from 4.0% (95% CI 3.5% to 4.6%) to 2.9% (2.6% to 3.2%), and of jaundice for pancreatic cancer from 12.8% (7.3% to 21.6%) to 6.3% (4.5% to 8.7%). Adding hidden text records did not alter the PPV of abdominal pain for bladder (codes: 0.14%, 0.13% to 0.16% vs codes plus hidden text: 0.14%, 0.13% to 0.15%) or pancreatic (0.23%, 0.21% to 0.25% vs 0.21%, 0.20% to 0.22%) cancer. Conclusions Omission of text records from CPRD studies introduces bias that inflates outcome measures for recognised alarm symptoms. This potentially reinforces clinicians’ views of the known importance of these symptoms, marginalising the significance of ‘low-risk but not no-risk’ symptoms.

Comorbid conditions delay diagnosis of colorectal cancer: a cohort study using electronic primary care records

Background:Pre-existing non-cancer conditions may complicate and delay colorectal cancer diagnosis.Method:Incident cases (aged ⩾40 years, 2007–2009) with colorectal cancer were identified in the Clinical Practice Research Datalink, UK. Diagnostic interval was defined as time from first symptomatic presentation of colorectal cancer to diagnosis. Comorbid conditions were classified as ‘competing demands’ (unrelated to colorectal cancer) or ‘alternative explanations’ (sharing symptoms with colorectal cancer). The association between diagnostic interval (log-transformed) and age, gender, consultation rate and number of comorbid conditions was investigated using linear regressions, reported using geometric means.Results:Out of the 4512 patients included, 72.9% had ⩾1 competing demand and 31.3% had ⩾1 alternative explanation. In the regression model, the numbers of both types of comorbid conditions were independently associated with longer diagnostic interval: a single competing demand delayed diagnosis by 10 days, and four or more by 32 days; and a single alternative explanation by 9 days. For individual conditions, the longest delay was observed for inflammatory bowel disease (26 days; 95% CI 14–39).Conclusions:The burden and nature of comorbidity is associated with delayed diagnosis in colorectal cancer, particularly in patients aged ⩾80 years. Effective clinical strategies are needed for shortening diagnostic interval in patients with comorbidity.

Predicting risk after aneurysm surgery.

Carlisle et al. raise an important subject in estimating the impact of surgery on long-term survival [1]. Information that helps patients to decide whether to consent to surgery is valuable. Their risk calculator assumes that survival depends on baseline mortality from Office for National Statistics data, modified by coefficients for a large number of factors that are assumed to predict mortality. These coefficients have each been estimated in other studies. Therefore, they are only valid under the constraints of the original models, where they are adjusted to allow for the effects of, and dependence between, the other explanatory variables. One cannot assume that they are either appropriate or hold true in the setting of the Carlisle risk calculator, so validation of the method is indeed required. The Carlisle risk calculator also uses risk coefficients obtained from cardiopulmonary exercise testing and assumes that these are independent of the risks due to underlying co-morbidities. It is unlikely that this is correct. For example, in a patient with reduced exercise capacity due to heart failure, the increase in risk should be obtained from either the heart failure or the reduced exercise capacity, but not both. Furthermore, by ‘importing’ the risks from other studies, it is not possible to calculate errors on the coefficients as applied in the Carlisle risk calculator. In practice, any predictive model can only provide a range of values, some of which are more probable than others. Errors are essential to appreciating this degree of uncertainty. For discussion of risk estimates with individual patients to be meaningful, knowledge of this degree of uncertainty is required. Without this, patients are unable to assess correctly their individual balance between risk and benefit.

Identifying clinical features in primary care electronic health record studies: methods for codelist development

Objective Analysis of routinely collected electronic health record (EHR) data from primary care is reliant on the creation of codelists to define clinical features of interest. To improve scientific rigour, transparency and replicability, we describe and demonstrate a standardised reproducible methodology for clinical codelist development. Design We describe a three-stage process for developing clinical codelists. First, the clear definition a priori of the clinical feature of interest using reliable clinical resources. Second, development of a list of potential codes using statistical software to comprehensively search all available codes. Third, a modified Delphi process to reach consensus between primary care practitioners on the most relevant codes, including the generation of an ‘uncertainty’ variable to allow sensitivity analysis. Setting These methods are illustrated by developing a codelist for shortness of breath in a primary care EHR sample, including modifiable syntax for commonly used statistical software. Participants The codelist was used to estimate the frequency of shortness of breath in a cohort of 28 216 patients aged over 18 years who received an incident diagnosis of lung cancer between 1 January 2000 and 30 November 2016 in the Clinical Practice Research Datalink (CPRD). Results Of 78 candidate codes, 29 were excluded as inappropriate. Complete agreement was reached for 44 (90%) of the remaining codes, with partial disagreement over 5 (10%). 13 091 episodes of shortness of breath were identified in the cohort of 28 216 patients. Sensitivity analysis demonstrates that codes with the greatest uncertainty tend to be rarely used in clinical practice. Conclusions Although initially time consuming, using a rigorous and reproducible method for codelist generation ‘future-proofs’ findings and an auditable, modifiable syntax for codelist generation enables sharing and replication of EHR studies. Published codelists should be badged by quality and report the methods of codelist generation including: definitions and justifications associated with each codelist; the syntax or search method; the number of candidate codes identified; and the categorisation of codes after Delphi review.

Does the GP method of recording possible cancer symptoms reflect the probability that cancer is present

Special Issue: Abstracts of the Cancer and Primary Care Research International (Ca-PRI) Network 8th Annual Meeting, 20-22 May 2015, Denmark

The risk of bladder cancer with non-visible haematuria: a primary care study using electronic records

S OF ORAL PRESENTATIONS