Commonly used definitions in real-world studies may underestimate the prevalence of renal disease among nonvalvular atrial fibrillation patients

In total, 15,860 patients were included. The median age was 77 years (IQR: 14) and 7134 (45.0%) of the patients were female. A detailed description of the demographic characteristics of the apixaban and warfarin patients in this population have been previously published [23]. The cumulative prevalence of the comorbidities can be seen in Figure 2.

Using a 1-year look back window, the three most common comorbidities were chronic pulmonary disease (19.2%), diabetes with chronic complications (15.5%) and congestive heart failure (14.7%). When applying a 3-year look back, chronic pulmonary disease and diabetes with chronic complications were still the most common conditions, at 21.7 and 18.0% respectively. However, the third most common condition changed to renal disease (17.9%), which increased in prevalence by a relatively large amount when applying the longer look back window (+5.9%). Using an ever look back window, the most common comorbidity was still chronic pulmonary disease (29.1%), followed by renal disease (27.6%). The third most common comorbidity was now cerebrovascular disease with at 21.7%.

The prevalence of all comorbidities was, as expected given the definitions, higher when using an ever look back window compared with one and 3-year periods (Figure 2). One exception to this was observed for diabetes without complications; the prevalence of this condition decreased slightly between the 1- and 3-year periods (-1.2%). The largest absolute increases in prevalence when applying the longer look back windows were observed for renal disease (+5.9 and +15.6% using a 3-year and ever look back, respectively), followed by malignancy (+2.5% and +11.4% for 3-year and ever look backs). The third largest increase was observed for cerebrovascular disease (+2.7%) when using a 3-year look back, and for chronic pulmonary disease (+9.9%) when using an ever look back window.

Median CCI scores did not vary markedly using different look back windows, increasing slightly from 0 (IQR = 0–2) to 1 (IQR = 0–2) and 1 (IQR = 0–3) using a 1 year compared with a 3 year and ever look back. The percentage of patients that had a CCI score >0 increased from 46.3% using a 1-year look back window to 69.0% using an ever look back (Figure 3).

In this study of treated NVAF patients in the UK, the burden of different comorbidities was high. The most common comorbidities varied depending on the look back window applied; when using an ever look back period chronic pulmonary disease was most common, followed by renal disease and cerebrovascular disease. The prevalence of all comorbidities apart from diabetes without complications increased with longer look back windows, with the largest increases observed for renal disease, malignancies, cerebrovascular disease and chronic pulmonary disease. In real-world data, short look back windows are often applied, either out of necessity or study design choice. They are particularly common in US insurance claims databases, where the rate of patient disenrollment can be high [25]. Our findings emphasize that when a shorter look back window is used, the prevalence of some chronic conditions may be underestimated. In our population of NVAF patients using UK CPRD/HES linked data; this was particularly marked for renal disease; as well as for malignancies, cerebrovascular disease and chronic pulmonary disease.

There are several possible explanations for the increase in prevalence of chronic conditions we observed when extending the look back window. In CPRD, diagnostic Read codes tend to be entered only once upon the initial diagnosis. It is possible that the extension of the look back window allows us to capture more initial instances of a diagnosis, particularly if follow-up visits are entered using less specific codes. This would have less of an impact for more severe patients treated in secondary care, as the relevant ICD-10 codes are re-entered in HES. For renal disease, it is also feasible that the patients with mild disease are seen relatively infrequently, as current guidelines only recommend frequent monitoring for more severe patients [26]. Other comorbidities, which were particularly affected by the increased look back window, were malignancies, cerebrovascular disease and chronic pulmonary disease. All of these conditions showed relatively modest increases in prevalence with the extension of the look back window from 1 to 3 year, at just under a 3% increase, but larger increases when extending the look back further to all available data. It is feasible that we are capturing some historical diagnoses from which patient may no longer require frequent care due to having completed treatment. The appropriateness of including these diagnoses when describing patient characteristics at a particular time point will depend on the study aim. Regarding the decrease in prevalence observed for diabetes without complications, this is likely due to the fact that more patients are characterized has having complications with the extension of the look back window, and consequently recategorized from diabetes without complication to diabetes with complications.

We are not alone in describing the impact of the look back period on prevalence estimates in pharmacoepidemiological studies [18,20,27–29]. Rassen et al. recently investigated the impact of varying look back windows on incidence and prevalence estimates in a UK (CPRD) and US (MarketScan) database, and described increased prevalence of 4.3- to 8.3- times higher in CPRD depending on the disease, and an 1.8- to 2.4-times higher prevalence in MarketScan [20]. This is in line with findings reported by Ekstrom and colleagues, who found marked increases in the estimated prevalence, coupled with decreased incidence estimates, of different diseases when expanding the look back periods using Swedish registry data [30]. Although the impact of using a short look back window on absolute risk estimates is marked and consistently reported across the literature, the consequent impact of this on relative effect estimates appears more varied [21,27–29,31]. This is likely due to multiple factors, including the strength of any potential confounding path and whether the look back windows are differential or nondifferential between treatment groups [28].

Our study has a number of limitations. First, the comorbidities estimated in this study are only reflective of NVAF patients treated with OACs. However, the majority of NVAF patients are expected to be treated with OACs in the UK, and this number is increasing over time. A UK study estimated in 2015–2016 that 73.9% of NVAF patients initiated anticoagulation therapy at diagnosis [32]. Nonetheless, by restricting the population to treated NVAF patients certain patient groups may be under represented; specifically patients who are slightly younger, and those with fewer comorbidities [32]. This should be borne in mind as a limitation regarding the generalizability of our results.

Second, prevalence estimates using administrative data are sensitive to the coding practices within the relevant healthcare system, which may change over time. Although the validity of the CPRD is considered to be high [33], here are examples of diagnoses being under recorded in both CPRD and HES for certain conditions [34,35]. In the UK, the Quality and Outcomes Framework (QOF) has incentivized recording of CKD in primary care based on two estimated glomerular filtration rate (eGFR) measurements <60 ml/min/1.73 m² [36]. However, prior studies have found that a relatively large proportion of patients meeting biochemical criteria for CKD could lack appropriate Read Codes. A recent analysis of data from the National Chronic Kidney Disease Audit found that 30% of patients with biochemical CKD, as defined by creatinine measures, lacked a Read Code for Stage 3–5 CKD; however, this varied between 4 and 100% on a practice level [36]. Nonetheless, the percentage of patients who had codes for other renal disorders, dialysis or a kidney transplant was higher. The prevalence of renal disease in the ever look back period in our study (27.6%) is comparable with the range of renal disease prevalence in the NVAF population in the literature; including studies utilizing biochemical criteria (22.2–35.5%) [37–39]. This implies that the prevalence of renal disease may be more accurately estimated in CPRD/HES data when an ever look back is applied and codes indicating unspecific renal disease, as opposed to staged CKD, are used. To increase the accuracy of prevalence estimates of renal disease in this data, it might have been appropriate to identify renal disease using estimated glomerular filtration rate (eGFR) results tests, as adopted in previous real-world studies [37–39]. This is because eGFR tests are expected to be frequently recorded in the data, as physicians are recommended to conduct a test if patients experience a change in treatment or development of a comorbidity [40]. Recently published data [23] indicate that the use of diagnostic codes and a short look back window in isolation could underestimate the prevalence of renal dysfunction, compared with additionally using laboratory test results

Using treatments to identify conditions in addition to diagnostic codes might also have mitigated the impact of varying the look back window for some conditions, for example diabetes. However, for those conditions where the largest increases in prevalences were observed – renal disease and malignancies – treatment information would be difficult to utilize. For renal disease, this is because the relevant prescription drugs such as ACE-inhibitors have multiple indications, and chemotherapy treatments which could be used to indicate active treatment for a malignancy are not recorded within either CPRD or HES. Other limitations of our study include the potentially reduced generalizability due to HES linkage only being available for English practices and the potential for erroneous recording of data. We also acknowledge that the accuracy of the prevalence estimates is a direct reflection of the accuracy of the code lists, which, although based on a publication [24], have not been specifically validated within CPRD. Strengths of our study includes the large sample size and the primary and secondary care linkage, which allowed us to capture diagnoses in multiple settings.

Our study describes a high comorbidity burden among NVAF patients newly initiating treatment in England, with chronic pulmonary disease present among almost one in three patients and renal disease in more than one in four. The estimated prevalence of several conditions, most notably renal disease and malignancies, increased markedly when using an all available compared with a 1-year look back window. Our findings highlight the importance of utilizing longer look back windows when estimating the prevalence of chronic conditions in the NVAF population and emphasizes the high comorbidity burden among these patients.

All authors met the four standards for Authorship Credit as published in the Future Science Group Author Disclosure Form, and contributed sufficiently to the work to take responsibility for the article content.

We would like to acknowledge the contribution of Robert Donaldson of Evidera, who assisted with the creation of the analytical files.

This study was sponsored by Bristol Myers-Squibb. S Ramagopalan and F Mehmud are employees of Bristol Myers-Squibb and hold stock options in Bristol Myers-Squibb. A Schultze, S Graham and B Nordstrom are salaried employees of Evidera and are prohibited from accepting remuneration of any sort from clients of Evidera. Evidera received funding from Bristol Myers-Squibb to conduct this study. The authors have no other relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript apart from those disclosed.

No writing assistance was utilized in the production of this manuscript.

The authors state that the study was conducted in accordance with the ethical principles that have their origin in the Declaration of Helsinki, that are consistent with the current guidelines for International Society for Pharmacoepidemiology (ISPE) Good Pharmacoepidemiology Practices (GPP), and in accordance with local applicable laws and regulations. The protocol and proposed conduct of cohort study were approved by the Independent Scientific Advisory Committee (ISAC; 17_188).

This work is licensed under the Attribution-NonCommercial-NoDerivatives 4.0 Unported License. To view a copy of this license, visit http://creativecommons.org/licenses/by-nc-nd/4.0/