Statistics from Altmetric.com
We thank the authors for their interest in our research1 which has attracted both professional and public media interest.2–4 Certainly, we accept that epidemiological studies can be at risk of ‘ecological fallacy’ and cannot adjust for all possible confounders. We make no claim that association proves causality. However, in offering a generic note of caution, Ford et al provide no specific hypothesis or insight to explain why the associations illustrated in our particular study could be explained by unacknowledged bias or an artefact of data aggregation. Taking the logic from their example, the counter hypothesis for our findings would be that people who are destined for a poor cancer outcome were selectively choosing to register at a general practice with a low gastroscopy rate—this seems unlikely.
We took great care to explore potential bias in our approach to data aggregation by undertaking extensive sensitivity analyses.5 Each case of oesophagogastric cancer was assigned a range of alternative variables to express the gastroscopy rate at their local practice. Hence, instead of expressing the rate as tertile groups alone, we examined quintiles, continuous variables (ie, the absolute rate for each individual practice) and age-specific rates for over 55-year-olds only. This meant reassigning practices and their relative ranking across a range of scenarios to produce alternative patient-level exposure variables. We studied, not one, but three measures of cancer outcome in patient-level binary logistic regression analyses. Although we cannot exclude ecological fallacy, the remarkable consistency of our results lends weight to our hypothesis and we expect most readers will accept that wide variation in referral practice could be linked to differences in cancer outcome.
The factors responsible for variation in gastroscopy rates between individual practice populations are likely to be complex, including differences in local symptom prevalence and ‘consultation behaviour’, and possible differences in GP access to local hospital-based gastroscopy services. However, the finding of wide variation within relatively small geographical areas (individual primary care trusts) suggests a role for unexplained variation. We have undertaken work to confirm that wide variation in gastroscopy rate exists between practices in close geographical proximity. Using our local endoscopy database and hospital information systems, we audited 13 082 consecutive gastroscopies (April 2009–March 2012) from 63 general practices that refer exclusively to our unit. As in our national study, the mean adjusted gastroscopy rate varied by more than twofold across low, middle and high tertile practices in our area. Consistent with differing referral thresholds, the yield of cancer for low, middle and high tertile groups was 2.2%, 1.6% and 1.0% of gastroscopies (p<0.01). Postcode mapping of these local practices confirmed ‘low’ and ‘high’ referring practices within the same square mile. This suggests a spectrum of practice.
Ford observes that cancer is rare as an underlying cause of dyspepsia, but this is not in dispute and, indeed, is an integral theme of our paper. In our discussion, we caution against a simplistic policy of widespread increases in gastroscopy and illustrate that this is unlikely to be affordable. Rather, we suggest that our findings could support a targeted approach at local level to identify a minority of practice populations with atypically low rates of investigation (‘local outliers’).
We are aware of the cost-effectiveness argument for ‘test-and-treat’ in simple dyspepsia in younger patients and allude to this approach already. However, we assume that their enthusiasm for ‘test-and-treat’ does not extend to older subjects at greater risk of cancer. We would point out that in our sensitivity analyses we found wide variation in general practice gastroscopy rates among the >55-year-old population (where non-invasive strategies are not recommended) and confirmed that belonging to a practice with a low rate of gastroscopy in older subjects was associated with poor outcome.
The concluding paragraph suggests that Ford et al may not have fully appreciated the focus of our study since their prescription for improvement in cancer outcome makes no mention of measures to identify and reduce variation in existing practice. The magnitude of variation we have shown in rates of gastroscopy is consistent with wide differences in clinical practice across primary care and possible inequality in access to investigation. It would be naive to think that variation of this order can be explained simply by differences in local consultation behaviour or rates of dyspepsia rather than genuine differences in clinical decision making or threshold for referral. The higher diagnostic yield of serious pathology observed for gastroscopies from low-referring practices surely gives this notion face validity. By showing an association between variation in investigation rates and cancer outcome, we have identified a strong reason to reflect on whether current guidelines are fit for purpose or implemented effectively. Guidelines are intended to reduce variation in practice.
Competing interests None.
Patient consent Obtained.
Provenance and peer review Not commissioned; internally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.