Abstract
This retrospective study uses a Canadian reporting system to validate an algorithm that identifies oral anticoagulation prescriptions given to patients with atrial fibrillation treated in emergency departments.
Introduction
Atrial fibrillation is seen commonly in emergency departments (EDs).1,2 Initiation of oral anticoagulation to patients treated in the ED has been associated with higher long-term use than when prescribing is left to outpatient care following discharge.3
Current data sets do not identify ED prescriptions. The ability to identify these prescriptions in large data sets could facilitate future studies aimed at increasing such prescriptions and could support long-term monitoring of ED oral anticoagulant prescription patterns.
Methods
This retrospective cohort study sought to validate an algorithm that identifies ED oral anticoagulant prescription provision for patients with atrial fibrillation using administrative health data sets. Using the National Ambulatory Care Reporting System,4 we identified patients who had visited 1 of 20 Ontario EDs with a primary diagnosis of atrial fibrillation, in the years 2009 through 2014. We excluded patients who would not benefit from5 or who had been taking oral anticoagulant agents within 90 days, were younger than 65 years (because we do not have comprehensive outpatient medication data for them), or had been admitted to the hospital. Data were abstracted from patient medical records for the first visit per patient. Using unique encoded identifiers, we linked the medical record data to databases held at ICES. The study was approved by the Research Ethics Board of Sunnybrook Health Sciences Centre. This report follows the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.
The reference standard was medical record documentation of provision of an oral anticoagulant prescription by any physician (emergency or consultant). Thirty-two algorithms selected a priori were tested. We varied the algorithms in the following ways: (1) including oral anticoagulant prescription fills that occurred on the same day (day 0) as the patient left the ED (because some patients are sent to the ED by another clinician who may have prescribed the agent); (2) counting the number of days following discharge that patients filled a prescription; (3) excluding prescription fills that occurred after an outpatient visit with a relevant clinician (because that clinician may have written the prescription being filled); (4) excluding patients with a history of venous thromboembolism; and (5) removing patients who had seen a relevant clinician 30 days prior to the emergency visit. We calculated the sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV), and 95% (CIs). All analyses were conducted with SAS statistical software version 9.3 (SAS Institute Inc).
Results
Of 2015 qualifying patients, 65% had a CHADS2 (congestive heart failure, hypertension, age >75 years, diabetes, and stroke) score of 2 or higher. Emergency department physicians prescribed warfarin for a median of 7 days and oral anticoagulant agents for a median of 30 days. Inclusion of day 0 substantially improved sensitivity, eg, 34% of patients analyzed in algorithm 3, prescriptions filled between days 1 and 3, to 83% of patients analyzed in algorithm 2, prescriptions filled between days 0 and 3 (Table 1). Using a longer follow-up period for prescription fills lowered specificity (91% in algorithm 2 to 74% in algorithm 10, prescriptions filled between days 0 and 30, and only slightly improved sensitivity (83% to 88%).
Table 1. Algorithm Test Characteristics Among Patients Treated in the Emergency Department.
| Algorithm No. | Days in Which Oral Anticoagulant Prescription Filled | True Positive | True Negative | False Positive | False Negative | Sensitivity (95% CI) |
Specificity (95% CI) |
Positive Predictive Value (95% CI) |
Negative Predictive Value (95% CI) |
Rate, % |
|---|---|---|---|---|---|---|---|---|---|---|
| Filled | ||||||||||
| 1 | 0 | 198 | 1576 | 36 | 205 | 0.491 (0.442-0.540) | 0.978 (0.971-0.985) | 0.846 (0.800-0.892) | 0.885 (0.870-0.900) | 11.6 |
| 2 | 0-3 | 335 | 1466 | 146 | 68 | 0.831 (0.794-0.868) | 0.909 (0.895-0.923) | 0.696 (0.655-0.737) | 0.956 (0.946-0.966) | 23.9 |
| 3 | 1-3 | 137 | 1502 | 110 | 266 | 0.340 (0.294-0.386) | 0.932 (0.920-0.944) | 0.555 (0.493-0.617) | 0.850 (0.833-0.867) | 12.3 |
| 4 | 0-4 | 338 | 1441 | 171 | 65 | 0.839 (0.803-0.875) | 0.894 (0.879-0.909) | 0.664 (0.623-0.705) | 0.957 (0.947-0.967) | 25.3 |
| 5 | 1-4 | 140 | 1477 | 135 | 263 | 0.347 (0.301-0.393) | 0.916 (0.902-0.930) | 0.509 (0.450-0.568) | 0.849 (0.832-0.866) | 13.7 |
| 6 | 0-7 | 346 | 1374 | 238 | 57 | 0.859 (0.825-0.893) | 0.852 (0.835-0.869) | 0.592 (0.552-0.632) | 0.960 (0.950-0.970) | 29.0 |
| 7 | 1-7 | 148 | 1410 | 202 | 255 | 0.367 (0.320-0.414) | 0.875 (0.859-0.891) | 0.423 (0.371-0.475) | 0.847 (0.830-0.864) | 17.4 |
| 8 | 0-14 | 350 | 1297 | 315 | 53 | 0.868 (0.835-0.901) | 0.805 (0.786-0.824) | 0.526 (0.488-0.564) | 0.961 (0.951-0.971) | 33.0 |
| 9 | 1-14 | 152 | 1333 | 279 | 251 | 0.377 (0.330-0.424) | 0.827 (0.809-0.845) | 0.353 (0.308-0.398) | 0.842 (0.824-0.860) | 21.4 |
| 10 | 0-30 | 356 | 1189 | 423 | 47 | 0.883 (0.852-0.914) | 0.738 (0.717-0.759) | 0.457 (0.422-0.492) | 0.962 (0.951-0.973) | 38.7 |
| 11 | 1-30 | 158 | 1225 | 387 | 245 | 0.392 (0.344-0.440) | 0.760 (0.739-0.781) | 0.290 (0.252-0.328) | 0.833 (0.814-0.852) | 27.1 |
| Filled Prior to Any Appointment With a Family Physician, Cardiologist, or Internist | ||||||||||
| 12 | 0-14 | 264 | 1560 | 52 | 139 | 0.655 (0.609-0.701) | 0.968 (0.959-0.977) | 0.835 (0.794-0.876) | 0.918 (0.905-0.931) | 15.7 |
| 13 | 0-30 | 272 | 1557 | 55 | 131 | 0.675 (0.629-0.721) | 0.966 (0.957-0.975) | 0.832 (0.791-0.873) | 0.922 (0.909-0.935) | 16.2 |
| 14 | 0-30a | 272 | 1557 | 55 | 131 | 0.675 (0.629-0.721) | 0.966 (0.957-0.975) | 0.832 (0.791-0.873) | 0.922 (0.909-0.935) | 16.2 |
| Filled Prior to Any Appointment With a Family Physician, Cardiologist, or Internist and Had No Appointments 30 d Prior to ED Visit | ||||||||||
| 15 | 0-30 | 98 | 1597 | 15 | 305 | 0.243 (0.201-0.285) | 0.991 (0.986-0.996) | 0.867 (0.804-0.930) | 0.840 (0.824-0.856) | 5.6 |
| Filled Prior to Any Appointment With Their Family Physician, Cardiologist, or Internist | ||||||||||
| 16 | 0-30 | 236 | 1567 | 45 | 167 | 0.586 (0.538-0.634) | 0.972 (0.964-0.980) | 0.840 (0.797-0.883) | 0.904 (0.890-0.918) | 14.0 |
Abbreviation: ED, emergency department.
Includes a nephrologist.
After restricting to patients who filled a prescription before they saw another clinician, the number of false-positive results (ie, prescriptions filled that were not written in the ED) decreased. Thus, specificity improved from 74% in algorithm 10 to 97% in algorithm 13; however, sensitivity fell from 88% to 68%. Removal of patients with an outpatient visit prior to the ED visit (eg, algorithm 15) further increased specificity (99%) but decreased sensitivity (24%). Excluding patients with a prior venous thromboembolism resulted in minimal change (Table 2).
Table 2. Algorithm Test Characteristics Among Patients Treated in the Emergency Department With No Venous Thromboembolism Diagnosis in the Previous 5 Years.
| Algorithm No. | Days in Which Oral Anticoagulant Prescription Filled | True- Positive | True Negative | False Positive | False Negative | Sensitivity (95% CI) |
Specificity (95% CI) |
Positive Predictive Value (95% CI) |
Negative Predictive Value (95% CI) |
Rate, % |
|---|---|---|---|---|---|---|---|---|---|---|
| Filled | ||||||||||
| 17 | 0 | 194 | 1524 | 35 | 197 | 0.496 (0.446-0.546) | 0.978 (0.971-0.985) | 0.847 (0.800-0.894) | 0.886 (0.871-0.901) | 11.7 |
| 18 | 0-3 | 325 | 1418 | 141 | 66 | 0.831 (0.794-0.868) | 0.910 (0.896-0.924) | 0.697 (0.655-0.739) | 0.956 (0.946-0.966) | 23.9 |
| 19 | 1-3 | 131 | 1453 | 106 | 260 | 0.335 (0.288-0.382) | 0.932 (0.920-0.944) | 0.553 (0.490-0.616) | 0.848 (0.831-0.865) | 12.2 |
| 20 | 0-4 | 328 | 1394 | 165 | 63 | 0.839 (0.803-0.875) | 0.894 (0.879-0.909) | 0.665 (0.623-0.707) | 0.957 (0.947-0.967) | 25.3 |
| 21 | 1-4 | 134 | 1429 | 130 | 257 | 0.343 (0.296-0.390) | 0.917 (0.903-0.931) | 0.508 (0.448-0.568) | 0.848 (0.831-0.865) | 13.5 |
| 22 | 0-7 | 336 | 1329 | 230 | 55 | 0.859 (0.825-0.893) | 0.852 (0.834-0.870) | 0.594 (0.554-0.634) | 0.960 (0.950-0.970) | 29.0 |
| 23 | 1-7 | 142 | 1364 | 195 | 249 | 0.363 (0.315-0.411) | 0.875 (0.859-0.891) | 0.421 (0.368-0.474) | 0.846 (0.828-0.864) | 17.3 |
| 24 | 0-14 | 340 | 1255 | 304 | 51 | 0.870 (0.837-0.903) | 0.805 (0.785-0.825) | 0.528 (0.489-0.567) | 0.961 (0.951-0.971) | 33.0 |
| 25 | 1-14 | 146 | 1290 | 269 | 245 | 0.373 (0.325-0.421) | 0.827 (0.808-0.846) | 0.352 (0.306-0.398) | 0.840 (0.822-0.858) | 21.3 |
| 26 | 0-30 | 346 | 1150 | 409 | 45 | 0.885 (0.853-0.917) | 0.738 (0.716-0.760) | 0.458 (0.422-0.494) | 0.962 (0.951-0.973) | 38.7 |
| 27 | 1-30 | 152 | 1185 | 374 | 239 | 0.389 (0.341-0.437) | 0.760 (0.739-0.781) | 0.289 (0.250-0.328) | 0.832 (0.813-0.851) | 27.0 |
| Filled Prior to Any Appointment With a Family Physician, Cardiologist, or Internist | ||||||||||
| 28 | 0-14 | 257 | 1508 | 51 | 134 | 0.657 (0.610-0.704) | 0.967 (0.958-0.976) | 0.834 (0.792-0.876) | 0.918 (0.905-0.931) | 15.8 |
| 29 | 0-30 | 265 | 1505 | 54 | 126 | 0.678 (0.632-0.724) | 0.965 (0.956-0.974) | 0.831 (0.790-0.872) | 0.923 (0.910-0.936) | 16.4 |
| 30 | 0-30a | 265 | 1505 | 54 | 126 | 0.678 (0.632-0.724) | 0.965 (0.956-0.974) | 0.831 (0.790-0.872) | 0.923 (0.910-0.936) | 16.4 |
| Filled Prior to Any Appointment With a Family Physician, Cardiologist, or Internist and Had No Appointments 30 d Prior to ED Visit | ||||||||||
| 31 | 0-30 | 95 | 1544 | 15 | 296 | 0.243 (0.200-0.286) | 0.990 (0.985-0.995) | 0.864 (0.800-0.928) | 0.839 (0.822-0.856) | 5.6 |
| Filled Prior to Any Appointment With Their Family Physician, Cardiologist, or Internist | ||||||||||
| 32 | 0-30 | 229 | 1515 | 44 | 162 | 0.586 (0.537-0.635) | 0.972 (0.964-0.980) | 0.839 (0.795-0.883) | 0.903 (0.889-0.917) | 14.0 |
Abbreviation: ED, emergency department.
Includes a nephrologist.
Algorithm 2 maximized specificity (91%; 95% CI, 90%-92%) while maintaining reasonable sensitivity (83%; 95% CI, 79%-87%) (PPV, 70%; 95% CI, 66%-74%, NPV, 96%; 95% CI, 95%-97%). Algorithm 13, prescription filled prior to seeing another relevant clinician, provided high specificity (97%; 95% CI, 96%-98%).
Limitations include using medical record documentation of ED prescription provision as reference standard, and our results may not apply to patients younger than 65 years.
We derived several algorithms that identify ED oral anticoagulant prescriptions in a large health data set. Depending on project goals, an algorithm can be selected to optimize specificity, sensitivity, PPV, or NPV.
References
- 1.Atzema CL, Austin PC, Miller E, Chong AS, Yun L, Dorian P. A population-based description of atrial fibrillation in the emergency department, 2002 to 2010. Ann Emerg Med. 2013;62(6):-. doi: 10.1016/j.annemergmed.2013.06.005 [DOI] [PubMed] [Google Scholar]
- 2.McDonald AJ, Pelletier AJ, Ellinor PT, Camargo CA Jr. Increasing US emergency department visit rates and subsequent hospital admissions for atrial fibrillation from 1993 to 2004. Ann Emerg Med. 2008;51(1):58-65. doi: 10.1016/j.annemergmed.2007.03.007 [DOI] [PubMed] [Google Scholar]
- 3.Atzema CL, Jackevicius CA, Chong A, et al. Prescribing of oral anticoagulants in the emergency department and subsequent long-term use by older adults with atrial fibrillation. CMAJ. 2019;191(49):E1345-E1354. doi: 10.1503/cmaj.190747 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Canadian Institute for Health Information National Ambulatory Care Reporting System (NACRS) metadata. https://www.cihi.ca/en/types-of-care/hospital-care/emergency-and-ambulatory-care/nacrs-metadata. Published 2019. Accessed November 22, 2019.
- 5.Skanes AC, Healey JS, Cairns JA, et al. ; Canadian Cardiovascular Society Atrial Fibrillation Guidelines Committee . Focused 2012 update of the Canadian Cardiovascular Society atrial fibrillation guidelines: recommendations for stroke prevention and rate/rhythm control. Can J Cardiol. 2012;28(2):125-136. doi: 10.1016/j.cjca.2012.01.021 [DOI] [PubMed] [Google Scholar]
