Validation and characterization of venous thromboembolism diagnoses in the Swedish National Patient Register among patients with rheumatoid arthritis

Objective To assess the validity of venous thromboembolism (VTE) diagnoses registered in the Swedish National Patient Register (NPR) in patients with rheumatoid arthritis (RA). Method We performed a validation study using manual chart reviews to validate ICD-10 codes for VTE from the NPR. We took a random sample of 269 VTE events registered at hospitals in Region Stockholm from 2009 to 2018 in patients with RA. Results Medical records for all 269 VTE events were available for review. Overall, the positive predictive value (PPV) for a VTE diagnosis was 95%. For incident VTE events, the PPV was 87% and ranged from 80% to 98% across six more or less restricted alternative definitions of incident VTE event. Out of 235 confirmed incident VTE events, the vast majority were diagnosed independently of the RA disease (three cases occurred as a result of clinical work-up for a presumed RA-related sign or symptom, and in 17 cases did the work-up involve a rheumatologist). Conclusions This study demonstrates high validity for VTE diagnoses recorded in the NPR for patients with RA, thus confirming that the NPR may be used to identify prevalent VTE as well as incident VTE events in patients with RA. Our results further demonstrate that in patients with RA, diagnoses of VTE are only marginally influenced by work-up related to the rheumatic disease, suggesting a modest impact of surveillance or diagnostic bias.

Venous thromboembolisms (VTEs), including deep vein thrombosis (DVT) and pulmonary embolism (PE), are common and potentially lethal medical conditions (1)(2)(3). Individuals with rheumatoid arthritis (RA) are at increased risk of VTE (4)(5)(6)(7), at least partially because of a higher prevalence of traditional risk factors for VTE. In addition, inflammation is known to upregulate procoagulatory factors and cause endothelial damage, which is thought to increase the risk of VTE (8,9). We and others have demonstrated that in patients with RA, high disease activity is linked to an increased risk of VTE (10,11).
In recent years, there have been concerns over a potential increase in the risk of VTE among individuals treated with janus kinase (JAK) inhibitors. Clinical phase III trials and a postmarketing safety study have indicated an increased VTE risk in patients treated with tofacitinib, especially for the higher dose of 10 mg twice daily (12,13). This has led to both the European Medicines Agency and the US Food and Drug Administration issuing cautions for VTE events in these patients (14,15). It is currently not known whether this safety signal is due to a drug-related mechanism, or if there are other explanations, such as channelling bias or even chance.
Given the above, there is a need for more studies that assess various aspects of VTE in patients with RA. The validity of register-based International Classification of Diseases (ICD) codes for VTE is one of the cornerstones to ensure high quality in such studies. Previous studies, from different contexts and in different (general) populations, have reported a range of positive predictive values (PPVs) for registered VTE diagnoses (16)(17)(18)(19). To date, no study has assessed the validity of a register-based VTE diagnosis in individuals with RA. Except for an increased risk of VTE associated with the RA disease itself, there are grounds for incidental findings or misclassification of DVT in RA patients, such as ruptured Baker's cysts and local inflammation, both of which can mimic the symptoms of a DVT and lead to work-up unravelling a VTE. The aim of this study was therefore to assess the validity of VTE diagnoses based on ICD-10 codes recorded in the Swedish National Patient Register (NPR) for patients with RA.

Study design
We performed a validation and characterization study using manual chart review to validate ICD-10 codes for VTE from the Swedish NPR in a population-based random sample of patients with RA from 2009 to 2018. Ethical approval was granted by the Regional Ethics Committee, Stockholm, Sweden (d-nr: 2019-02585).

Setting
Sweden has a population of just over 10 million, of whom 2.4 million live in Region Stockholm. Swedish healthcare is publicly funded; the vast majority of patients with RA are treated by rheumatologists located at public hospital-based clinics. During the study period, both PE and DVT were generally diagnosed and treated at emergency departments and/or during inpatient care at public hospitals. In Region Stockholm, six hospitals serve the entire region with emergency care (Karolinska University Hospital, Stockholm South General Hospital, Danderyd Hospital, Saint Göran's Hospital, Södertälje Hospital, and Norrtälje Hospital).

Data sources
We identified patients with RA using the Swedish Rheumatology Quality Register (SRQ). The SRQ is a longitudinal clinical rheumatology register operated by the Swedish Society for Rheumatology, covering around 85-90% of all Swedish RA patients. We linked this RA cohort to other national population-based Swedish health registers. To identify registered ICD-10 codes for VTE, we used the Swedish NPR. This register contains all registered ICD codes from inpatient care since 1969, and from specialized outpatient care since 2001. Since 1987, the coverage has been reported to be greater than 99% (20). We also obtained information on death via the Cause of Death Register, which contains the main and contributory causes of death since 1969, and information on dispensed drugs from the Prescribed Drug Register, which contains information on all dispensed drugs since July 2005 (21).

Study population and sampling process
For individuals with RA (incident or prevalent) registered in the SRQ between 2006 and 2018 (n = 46 316), we identified all inpatient and outpatient visits recorded in the NPR during the same period that listed an ICD-10 code for VTE at one of the six hospitals in Region Stockholm (n = 10 726), and death from PE from the Cause of Death register (n = 36). Due to a scarcity of VTE events, Norrtälje Hospital was excluded. Among the first registered VTE event for each individual between 2009 and 2018 (n = 451), we took a random sample of 275 events. Of these, 269 events were recorded in the NPR, and thereby included in the chart validation. The remaining six events were deaths from PE, as listed in the Cause of Death Register. These events could not be validated via medical records since none of the RA patients was treated at any of the hospitals included in this study at, or near to, the time of the VTE, and we did not have access to autopsy reports. See Figure 1 for a further description of the sampling process.

Definition of VTE
Our main definition of VTE event included ICD-10 I26 (PE), or I80, I81, and I82 [DVT and superficial vein thrombosis (SVT)] as the main or first contributory diagnosis recorded in the NPR. We included first time as well as recurring VTEs. An event was considered incident if it occurred (according to the medical record) at or within 90 days prior to the date of the corresponding ICD-10 code in the NPR, or (for inpatient records) at any time during the hospital stay. An event was considered prevalent if it occurred (according to the medical record) more than 90 days prior to the date of registering the ICD-10 code in the NPR. A registration listing an ICD-10 code for PE was considered valid if confirmed by computed tomography (CT) scan or  scintigraphy. A registration listing an ICD-10 code for DVT was considered valid if confirmed by ultrasound, CT scan, or phlebography. SVT was included in the primary DVT definition, and was considered valid if confirmed by ultrasound or recorded without doubt by the treating physician. We validated the entire group of VTE ICD-10 codes above (incident and prevalent combined), as well as incident VTE only. For incident VTE, and in sensitivity analyses, we also applied the following restrictions where we calculated the PPV, and the associated loss in sensitivity (defined as the number of all confirmed incident VTE events included in the subset divided by the total number of confirmed incident VTEs): • Restriction 1: As main definition of incident VTE but excluding those individuals with a previous VTE registration in the NPR within the last year (to assess the impact of any misclassification related to our definition of prevalent vs incident VTE). • Restriction 2: As main definition of incident VTE but with the additional requirement of a filled prescription of an anticoagulant within 30 days after the VTE event [see Table S2 for Anatomical Therapeutic Chemical (ATC) classification codes], unless death occurred from any cause within 30 days. In addition, we defined VTE events according to each individual ICD-10 code within the main definition (I26 and I80.0-I82.9), and calculated the PPV and sensitivity for each code.

Chart review
Medical records were manually reviewed by a physician (VM) using a data extraction form (see Table S1) created for this study. In summary, the following parameters were extracted during the review: (i) whether the registered VTE could be confirmed as a VTE or not; (ii) whether the confirmed VTE was incident or prevalent; (iii) the clinical phenotype of the VTE; (iv) the diagnostic modality and findings supporting the VTE diagnosis; (v) the reason for the clinical work-up leading to the VTE diagnosis; (vi) whether a set of predefined VTE risk factors was present (obesity, immobilization, hospital admission, pregnancy, or puerperium); and (vii) the anti-rheumatic treatment (if any) ongoing at the time of the VTE. Through the register linkage described above, the chart review data were enriched with data on previous VTE, filled prescriptions for anticoagulants, and an additional set of VTE risk factors (recent joint surgery, cancer, coagulation disorder, previous VTE, and oral hormone therapy). Definitions of VTE risk factors and anti-rheumatic treatment are presented in Table S2.

Statistical analyses
For the main definition and for each of the restricted definitions, we calculated the PPV for VTE (the number of incident VTE events divided by the total number of VTE events for each definition), overall and separately for PE and DVT. The estimation of 95% confidence intervals assumed a normal approximation to the binomial distribution. Analyses were performed using SAS version 9.4.

Results
We were able to collect medical records including radiology reports for each of the 269 registered VTE events. Of these, 109 were registered at Karolinska University Hospital, 67 at Stockholm South General Hospital, 48 at Saint Göran's Hospital, 30 at Danderyd Hospital, and 15 at Södertälje Hospital.
Among the 269 patients with registered VTE events included for validation, the mean age at VTE was 71 years and 183 (68%) were female. In total, 255 events were confirmed as VTE (incident or prevalent), of which 235 were incident and 20 were prevalent. Of the 235 incident VTEs, 139 were DVT or SVT, and 96 were PE. For all of the 20 confirmed prevalent VTEs, the ICD-10 code for VTE was registered at a coagulation outpatient clinic during a follow-up visit after a previous VTE. Of the 14 unconfirmed VTEs, nine were registered owing to clinical suspicion of DVT where ultrasound (usually during the following day) turned out to be negative. In five of these cases, ultrasound revealed a ruptured Baker's cyst (Table 1).
For confirmed incident DVT and SVT, the diagnosis was based on findings from an ultrasound or a CT scan in 128 cases (92%). The remaining 11 events (all of which were SVTs) were based on clinical diagnosis. In total, 229 (97%) of the confirmed incident VTE events were found to be true VTE by the reviewer, whereas six (3%) were judged as somehow uncertain by the reviewer, although treated as VTE by the clinician. Moreover, 224 (97% of all incident VTEs) of the individuals with a confirmed VTE had symptoms resulting in clinical suspicion of VTE, but in eight cases (3% of all incident VTEs) the VTE diagnosis was the result of an incidental radiological finding. In three cases (1% of all incident VTEs), there was an initial suspicion of an RA-related cause for the symptoms rather than VTE. In 17 cases (7% of all incident VTEs), the VTE work-up was initiated by a rheumatologist suspecting a VTE at a regular outpatient visit.
For our main definition, and combining incident and prevalent VTE, the PPV was 95%. For our main definition of an incident VTE event, the PPV was 87%, with slightly higher PPV for PE (90%) compared to DVT (86%). In our five alternative (restricted) definitions of VTE, the PPV for incident VTE ranged from 80% to 98%, and the corresponding sensitivities ranged from 94% to 79% (Table 3). PPVs for individual ICD-10 codes are presented in Table S3.

Discussion
In this study, the first to assess the validity of registerbased ICD-10 codes for VTE specifically in individuals with RA, we found a high PPV for incident and  prevalent VTE (95%) as well as for incident VTE only (87%), which were similar for DVT and PE. When using stricter definitions of VTE, the PPV was even higher, although at the expense of up to 20% loss in sensitivity compared to our main definition. Previous studies validating registered ICD codes for VTE have demonstrated a wide range of PPVs, generally lower than those in the current study. For several reasons, it is difficult to compare results between validation studies. First, different studies use different ICD codes to define VTE, as well as different strategies of including codes for outpatient care in addition to codes from inpatient care. Secondly, the availability of better diagnostic modalities in recent years, in combination with improved accuracy when using ICD codes over time, is likely to generally increase PPV for more recent periods. Thirdly, clinical characteristics including VTE risk factors are usually not presented in these studies, whereas the PPV is directly proportional to the prevalence of the condition under study. Since our study included only individuals with RA (who are at 50-100% increased risk of VTE compared to the general population), the PPV is likely to be somewhat higher than in otherwise unselected individuals registered with a VTE.
A Swedish study assessing the validity of VTE in 2450 individuals with a first time VTE between 1985 and 2014 found PPVs for PE and DVT of 80.7% and 59.2%, respectively (16), both lower than in our study. There may be a couple of reasons for this. First, the study period differs from that of the current study. As mentioned above, this could be an explanation for the higher validity observed in our study. Secondly, the previous study used VTE events from a general population cohort included in a health screening programme, aged 30-60 years at inclusion. Thus, compared to our study population, there is a discrepancy in the underlying risk of VTE and occurrence of VTE risk factors. A Danish study validated a random sample of 100 cases of VTE from 2010 to 2012 in the Danish National Patient Registry (17), a register similar to the one used in the current study. That study found a PPV of 88% for first time VTE and 72% for recurring VTE, with small differences between PE and DVT. Although covering a more recent period than the Swedish study (16), the Danish study (17) used VTE events from the general population and included a narrower set of ICD codes to define DVT, which may explain some of the discrepancy with the results in our study.
When excluding individuals with a prior registered VTE event within the last year, in an attempt to remove prevalent cases, there was an increase in the PPV for incident VTE, at the expense of some loss of sensitivity. Applying the requirement of filled prescription for an anticoagulant (or death from any cause) within 30 days resulted in a more distinct increase in PPV. Combining these two restrictions resulted in the highest PPV (98%). Our results thus suggest that applying this combination of restrictions results in a very high PPV for incident VTE when including both first time and recurring ICD codes for VTE. At the same time, these restrictions come at the expense of a loss in sensitivity, i.e. not capturing all confirmed incident VTEs compared to when not using the restriction. This finding illustrates that the 'best' approach to define VTE will depend on the exact research question to be tested. In our main analysis, we included the entire spectrum of ICD-10 codes for VTE, except for those specific to VTE related to pregnancy. This was to enable us to investigate the distribution of clinical phenotypes and PPV for each of the individual ICD-10 codes for VTE. In doing so, we noticed some misclassification. For example, among the 20 cases of SVT in this study, the majority were distributed within the ICD-10 codes commonly used for DVT. One explanation for this misclassification could be the slightly unspecific and somewhat overlapping code descriptions for VTEs in the ICD-10 system.
For (only) 17 of the confirmed VTE cases was a rheumatologist involved in the work-up. Only three confirmed VTE events presented in a way that was initially interpreted as a symptom of the RA disease. This indicates that RA itself had little impact on the clinical work-up and diagnosis of VTE, and implies that there is no major degree of surveillance or diagnostic bias of VTE within the RA population, thus enabling comparison of the VTE risk with the general population.
A limitation of this study is that, for administrative reasons, it was limited to hospitals in Region Stockholm. However, since these hospitals cover almost the entire region, which includes around 23% of the Swedish population, covering all socioeconomic subsets of the general population, this is not likely to affect the generalizability of these results.
The study has several strengths. Using the Swedish Rheumatology Register to identify RA patients, as well as using the NPR to identify VTE events, minimized selection bias and increased generalizability owing to the high coverage of these registers. Also, using other national health registers for additional clinical data, such as filled prescription of anticoagulants, offered a unique possibility to test alternative definitions of VTE. We were also able to manually review medical records for 100% of the selected VTE events.

Conclusion
Based on a review of inpatient and specialized outpatient medical records, the validity of registered ICD-10 codes for VTE was high, with a PPV of 95%, and a PPV for incident VTE events of 87%. These results verify that the Swedish NPR can be used to identify VTE events with high validity in individuals with RA, and that VTE in RA is mostly diagnosed on its own merits rather than on the basis of clinical work-up related to the RA disease.