Introduction: CHA2DS2‐VASc and CHADS2 are computationally simple risk prediction tools used to guide anticoagulation decisions for stroke prophylaxis, but they have modest risk discrimination ability and use static dichotomous variables. The Intermountain Mortality Risk Scores (IMRS) are dynamic decision tools using standard clinical laboratory tests. This study derived new stroke prediction scores using variables from both CHA2DS2‐VASc and IMRS. Methods and Results: In outpatients with first atrial fibrillation (AF) diagnosis at the Intermountain Healthcare (females, n = 26 063 males, n = 29 807), sex‐specific "IMRS‐VASc" scores were derived using variables from CHA2DS2‐VASc, warfarin use, the complete blood count, and the comprehensive metabolic profile. Validation was performed in an independent Intermountain outpatient AF cohort (females, n = 11 021; males, n = 12 641). Stroke occurred among 3.1% and 3.1% of females and 2.3% and 2.5% of males in derivation and validation groups, respectively. IMRS‐VASc stratified stroke with similar ability in derivation (c‐statistics, females: c = 0.703, males: c = 0.697) and validation groups (females: c = 0.681, males: c = 0.685). CHA2DS2‐VASc (females: c = 0.581 and c = 0.605; males: c = 0.616 and c = 0.613 in derivation and validation, respectively) and CHADS2 (females: c = 0.581 and c = 0.608; males: c = 0.620 and c = 0.621 in derivation and validation, respectively) were substantially weaker stroke predictors. IMRS was the strongest mortality predictor (females: c = 0.783 and c = 0.782; males: c = 0.796 and c = 0.794 in derivation and validation, respectively) and all scores were poor at predicting bleeding risk. Conclusions: A temporally dynamic risk score, IMRS‐VASc was derived and validated as a predictor of stroke in outpatients with AF. IMRS‐VASc requires further validation and the evaluation of its use in guiding care and treatment decisions for patients with AF.
Keywords: clinical decision tool; IMRS; IMRS‐VASc; learning health system; risk stratification
Atrial fibrillation (AF) is associated with systemic comorbidities, conduction abnormalities, and a five‐fold increase in thromboembolism risk.[
A more dynamic risk stratification tool that provides information regarding the need for OAC and other care or treatment decisions would be beneficial in clinical practice. Such a tool could consider factors that are not only commonly available across the continuum of care but that are dynamic—increasing or decreasing based on changes in a patient's health or disease management, or changes in lifestyle choices that may alter long‐term AF‐related comorbidities. The Intermountain Mortality Risk Scores (IMRS) are augmented intelligence decision tools composed using parameters from the complete blood count (CBC) and basic metabolic profile (BMP) and have been extensively validated to stratify morbidity, and mortality in a variety of patient populations in various geographic locations.[[
This study evaluated the potential combination of risk factors from CHA
The objective of the study was to develop and validate sex‐specific aggregate risk metrics that provide greater ability to predict incident stroke among patients with AF than previously developed risk scores. This risk score derivation was proposed to be conducted using the commonly available and familiar components of the CHA
1 Baseline characteristics by sex in the validation and derivation groups
Females Males Validation Derivation Validation Derivation Age, y 73.0 ± 13.2 73.3 ± 12.9 69.0 ± 13.7 69.1 ± 13.8 Hypertension 62.8% 63.4% 57.7% 58.3% Hyperlipidemia 39.6% 39.1% 42.6% 43.3% Diabetes 24.0% 23.9% 25.1% 25.3% Smoking 14.5% 14.9% 28.3% 28.3% CAD 28.3% 28.2% 41.8% 41.7% Prior MI 7.3% 6.9% 11.4% 11.2% Heart failure 30.2% 30.0% 26.8% 26.6% Prior stroke/TIA 9.8% 9.8% 7.8% 7.5% PVD 4.5% 4.5% 5.5% 5.1% Prior pulmonary embolism 4.5% 4.4% 3.3% 3.6% Renal failure 10.7% 10.9% 12.9% 12.5% Sleep apnea 10.7% 11.1% 16.1% 16.5% Cardiomyopathy 5.1% 5.0% 7.1% 7.1% Prior malignancy 10.0% 10.4% 12.9% 13.5% CHADS2 ≤1 44.2% 44.6% 53.1% 52.5% 2 30.5% 29.9% 26.5% 27.2% ≥3 25.4% 25.4% 20.5% 20.3% CHA2DS2‐VASc ≤1 8.8% 9.1% 28.5% 28.4% 2 12.6% 12.5% 21.2% 21.1% ≥3 78.6% 78.4% 50.3% 50.5% IMRS Low 20.3% 19.6% 35.1% 34.4% Moderate 47.1% 47.3% 45.4% 45.4% High 32.6% 33.2% 19.5% 20.2% ACE inhibitor 13.9% 14.4% 15.6% 15.7% ARB 7.3% 7.5% 5.3% 5.4% Antiarrhythmic 2.3% 2.2% 3.1% 3.2% Beta‐blocker 17.0% 16.8% 18.1% 18.4% CCB 11.1% 11.5% 9.6% 9.5% Diuretic 21.5% 21.8% 18.9% 18.5% Statin 13.8% 13.5% 17.5% 17.5% Aspirin 19.2% 19.0% 21.8% 21.9% Warfarin 11.1% 11.1% 9.2% 9.2%
1 Abbreviations: ACE, Angiotensin converting enxyme ihibitor; ARB, angiotensin II receptor blockers; CAD, coronary artery disease; CCB, calcium channel blocker; IMRS, intermountain mortality risk scores; MI, myocardial infarction; PVD, peripheral vascular disease; TIA, transient ischemic attack.
2 * IMRS categorizations were made based on previously‐derived sex‐specific thresholds.[
This study evaluated N = 79 532 (n = 37 084 females, and n = 42 448 males), community‐dwelling patients with non‐valvular AF who were ≥18 years of age and clinically managed at Intermountain Healthcare. The study included patients seen from January 1990 to May 2013. The index date of the study for each patient was the date of their first AF diagnosis. Only patients with an available CBC or an available BMP/CMP, or both, from the time of or within 6 months before the initial AF diagnosis were included in the study, with the laboratory results closest to but not after the initial AF diagnosis being used in the case that multiple CBC or metabolic profile panels in that time window were available. IMRS‐VASc scores were derived in 70% of this population (n = 26 063 females; n = 29 807 males) to identify the significantly predictive variables and determine weightings (see below). These IMRS‐VASc scores were then validated using the other 30% of the population (n = 11 021 females; n = 12 641 males) that had been held aside for testing, whether the derived scores were predictive in an independent population. This study was approved by the Intermountain Healthcare Institutional Review Board as a data‐only study with a waiver of consent because it posed minimal risk to subjects.
The IMRS‐VASc derivation was generated based on the processes used for the derivation of IMRS previously, but the weightings derived here for IMRS‐VASc are completely different from those in the standalone IMRS that was previously calibrated for mortality.[
IMRS‐VASc also added the components of the CHA
The IMRS‐VASc scoring system was designed from sex‐specific multivariable Cox regression models in the derivation population that initially considered each factor from the CBC panel, each parameter of the CMP, each component used in CHA
Mortality was determined from electronic records of all 22 Intermountain hospitals, from electronic Utah death certificates, and from the social security death master file. Utah death certificates and hospital records provided causes of death for the majority of deceased patients. Incidence of stroke and bleeding were established using International Classification of Disease coding (ICD‐9 and ICD‐10) from the Intermountain electronic data warehouse that contains data from Intermountain hospitals and more than 180 clinics. Intermountain provides healthcare services to approximately two‐thirds of the population of Utah and southeastern Idaho. All patients had at least 5‐years of follow‐up. For survival analyses examining stroke outcomes, patients were censored at 2 years after entry into the study if they had no fatal or nonfatal stroke and were censored as non‐events before 2 years if the patient passed away due to a non‐stroke cause. Secondary time points were also examined at 3 and 5 years. IMRS‐VASc was categorized for clinical relevancy of analyses into tertiles, with the tertiles 1, 2, and 3 being considered to be low, moderate, and high‐risk groups, respectively.
A secondary bleeding outcome was also examined in the 2‐, 3‐, and 5‐year time periods in the derivation population using censoring at those years or censoring for mortality as a non‐event as in the stroke analyses. Survival analyses were also performed examining all‐cause mortality. For bleeding and mortality, IMRS‐VASc scores were applied to the analyses as derived for the stroke endpoint, and IMRS,[
The derived IMRS‐VASc scores were applied for validation purposes to the independent set of 30% of patients that were held aside at the beginning of the study. Cox regression was used with IMRS‐VASc as the independent variable and stroke as the dependent variable. Both categorical and continuous IMRS‐VASc variables were evaluated and hazard ratios (HR), 95% confidence intervals (CI), and P values were computed. Kaplan‐Meier survival curves were used to graphically represent the survival estimates by risk category and the log‐rank statistic was used to examine significance. The stroke endpoint was examined with up to 2‐, 3‐, and 5‐year events, and bleeding and mortality outcomes were also examined in those time frames.
To compare the predictive capability of IMRS‐VASc risk scores to other scores, the IMRS, CHADS
Baseline characteristics of females and males in the derivation and validation populations are shown in Table. IMRS‐VASc had the following ranges of integer risk score values when categorized into tertiles: tertile 1: ≤5, tertile 2: 6 to 7, tertile 3: ≥8. In both the derivation and validation populations, IMRS‐VASc had a greater discrimination ability for stroke than IMRS, CHADS
2 C‐statistics with 95% confidence intervals for 2‐y outcomes of the primary (2‐y) IMRS‐VASc and other risk scores for discrimination of stroke, mortality, and bleeding risk
Females 2‐y stroke 2‐y mortality 2‐y bleed IMRS‐VASc Derivation 0.706 (0.687, 0.726) 0.607 (0.599, 0.615) 0.559 (0.549, 0.570) Validation 0.681 (0.651, 0.711) 0.611 (0.598, 0.623) 0.561 (0.545, 0.577) IMRS Derivation 0.550 (0.536, 0.573) 0.783 (0.777, 0.790) 0.572 (0.562, 0.583) Validation 0.550 (0.521, 0.579) 0.782 (0.773, 0.792) 0.567 (0.551, 0.582) CHADS2 Derivation 0.581 (0.562, 0.599) 0.693 (0.686, 0.701) 0.590 (0.580, 0.600) Validation 0.608 (0.580, 0.635) 0.703 (0.692, 0715) 0.594 (0.578, 0.610) CHA2DS2‐VASc Derivation 0.581 (0.562, 0.599) 0.696 (0.689, 0.704) 0.599 (0.589, 0.610) Validation 0.605 (0.579, 0.632) 0.702 (0.691, 0.713) 0.607 (0.591, 0.623) Males IMRS‐VASc Derivation 0.697 (0.675, 0.718) 0.638 (0.630, 0.646) 0.564 (0.554, 0.574) Validation 0.685 (0.652, 0.719) 0.628 (0.616, 0.640) 0.569 (0.554, 0.585) IMRS Derivation 0.557 (0.537, 0.577) 0.796 (0.790, 0.802) 0.597 (0.587, 0.606) Validation 0.544 (0.515, 0.572) 0.794 (0.785, 0.804) 0.595 (0.581, 0.610) CHADS2 Derivation 0.620 (0.600, 0.639) 0.702 (0.694, 0.709) 0.606 (0.596, 0.615) Validation 0.621 (0.591, 0.651) 0.694 (0.682, 0.705) 0.619 (0.605, 0.634) CHA2DS2‐VASc Derivation 0.616 (0.597, 0.635) 0.704 (0.697, 0.711) 0.619 (0.610, 0.628) Validation 0.613 (0.583, 0.643) 0.701 (0.690, 0.712) 0.625 (0.610, 0.640)
3 Abbreviation: IMRS, intermountain mortality risk scores.
IMRS‐VASc had stroke frequencies in the validation population that were very similar to those found in the derivation population and was highly predictive of the incidence of stroke at 2 years across categories of low, moderate, and high‐risk (Table). In contrast, high‐risk CHA
3 Frequency of stroke by categories defined by (A) 2‐y IMRS‐VASc risk score tertiles, and (B) CHA 2 DS 2 ‐VASc score categories
A) IMRS‐VASc Tertile 1 Tertile 2 Tertile 3 Females Overall (low‐risk) (moderate risk) (high‐risk) 2‐y Stroke ≤5 6‐7 ≥8 Derivation 3.1% 1.4% 2.5% 6.7% <.001 (151/10 744) (211/8559) (451/6760) Validation 3.1% 1.8% 2.5% 5.8% <.001 (84/4608) (89/3587) (165/2826) Males ≤5 ≥8 2‐y Stroke 6‐7 Derivation 2.3% 0.9% 1.9% 4.2% <.001 (93/10 123) (179/9 648) (421/10 036) Validation 2.5% 1.5% 1.6% 4.3% <.001 (62/4275) (65/4039) (187/4327)
3 Frequency of stroke by categories defined by (A) 2‐y IMRS‐VASc risk score tertiles, and (B) CHA 2 DS 2 ‐VASc score categories
B) CHA2DS2‐VASc ≤1 2 ≥3 Females Overall (low‐risk) (moderate risk) (high‐risk) 2‐y Stroke 1 2 ≥3 Derivation 3.1% 1.1% 1.9% 3.6% <.001 (26/2381) (62/3279) (725/20 403) Validation 3.1% 0.4% 1.6% 3.6% <.001 (4/972) (23/1398) (311/8651) Males ≤1 ≥3 2‐y Stroke 2 Derivation 2.3% 1.0% 2.0% 3.2% <.001 (87/8490) (124/6302) (482/15 015) Validation 2.5% 1.3% 2.0% 3.3% <.001 (48/3605) (54/2686) (212/6350)
4 Abbreviation: IMRS, intermountain mortality risk scores.
In secondary analyses of substrata defined by CHA
In secondary analyses of 2‐year mortality and 2‐year bleeding outcomes, IMRS‐VASc stratified risk of these outcomes in a highly statistically significant manner (Table). It was weaker, though, at discriminating mortality and bleeding risk than IMRS, CHADS
4 Frequency of mortality and bleeding by categories defined from tertiles of the 2‐y IMRS‐VASc risk scores
Overall Events Tertile 1 Tertile 2 Tertile 3 Females (low‐risk) (moderate risk) (high‐risk) 2‐y Mortality Derivation 21.0% 13.8% (1478/10 744) 23.1% (1976/8559) 29.8% (2014/6760) <.001 Validation 21.6% 15.1% (697/4608) 22.4% (805/3587) 31.0% (875/2826) <.001 2‐y Bleed Derivation 11.9% 9.3% (1004/10 744) 13.0% (1113/8559) 14.5% (983/6760) <.001 Validation 11.9% 9.3% (427/4608) 13.0% (465/3587) 14.8% (418/2826) <.001 Males 2‐y Mortality Derivation 19.2% 12.4% (1258/10 123) 18.8% (1810/9648) 26.5% (2659/10 036) <.001 Validation 18.9% 12.9% (553/4275) 18.6% (750/4039) 25.1% (1088/4327) <.001 2‐y Bleed Derivation 11.8% 8.7% (885/10 123) 12.0% (1154/9648) 14.7% (1472/10 036) <.001 Validation 11.4% 8.5% (362/4275) 11.3% (458/4039) 14.4% (623/4327) <.001
5 Abbreviation: IMRS, intermountain mortality risk scores.
The percentages of 3‐ and 5‐years stroke by 3‐ and 5‐years IMRS‐VASc scores are shown in Table S3. HRs and CIs for these scores and these time frames of stroke are provided in Figure. As representative results, the Kaplan‐Meier survival curves for 5‐year IMRS‐VASc in 5‐year stroke are shown in Figure S4. C‐statistics for 3‐ and 5‐years scores are provided in Table S4 for these secondary models. Finally, the association of 3‐ and 5‐year scores with mortality and bleeding outcomes are provided in Table S5. Sensitivity analyses are found in Table S6 for CHA
In adults with newly‐diagnosed non‐valvular AF, an augmented intelligence decision tool "IMRS‐VASc" was derived using component variables from CHA
AF is a prevalent and complex disease that reflects systemic disease as well as local adverse atrial remodeling.[[
Guideline‐based care utilizes the CHA
Despite these limitations, the CHA
In contrast, IMRS‐VASc improves the ability to predict stroke and, as with CHA
As in previous risk score applications,[[
An advantage of IMRS‐VASc is that it also performs well in females. Some component variables predicted stroke only in females, including sodium, bicarbonate, bilirubin, mean corpuscular hemoglobin concentration, and red cell distribution width, which may reveal important pathways related to stroke that apply most strongly to females and should be investigated further for possible sex differences in stroke risk. Further consideration clinically is needed for older females and other subpopulations that may be at elevated risk of bleeding or have other concerns that suggest a less aggressive approach to anticoagulation. Having improved ability to discern risk among females may aid in this effort.
This observational study utilized data previously collected in a prospective fashion among patients undergoing clinical care in a primary care setting. Due to its observational nature and the retrospectively chosen hypothesis, the population selection and data gathering may have resulted in unobserved or uncontrolled confounding. The study used disease classification coding to define prospective events and, although Intermountain serves more than two‐thirds of patients in the Utah and Southern Idaho geographies, some patients may have sought follow‐up care elsewhere. The study's approach to data collection may also suffer from missed events due to subclinical outcomes. Further, the time frame of patients studied may include secular trends due to changes in modalities of healthcare over time such as the use of new medications or devices, and evolving standards of care. The study's design may limit the generalizability of findings to other populations. Additional studies are needed to validate IMRS‐VASc in association with stroke in other populations with different racial compositions and distinct baseline stroke rates. If validated in other geographic locales, the evaluation of clinical use of the risk scores will require prospective implementation and evaluation.
Strengths of the study include a novel paradigm for risk scoring that uses practical, parsimonious models that may reduce the work of clinicians by employing the electronic health record for aiding physicians rather than adding burden in which they would need to gather data and calculate a score for each patient. Also, the use of several decades of data and the presence of stroke cases and of controls across each of those years should help balance any bias due to changing care processes. The primary components of the scores were common clinical laboratory tests that are ubiquitous in patient records and have been consistently defined and collected for over fifty years in medicine.
A newly‐derived risk score for predicting stroke, IMRS‐VASc was validated in association analyses with stroke outcomes among a second, independent set of outpatients with AF. It also was shown to have greater predictive ability than other risk scores, including standard risk scores used for the care of patients with AF. While it did predict mortality and bleeding outcomes, it was not as powerful at predicting those outcomes as IMRS (the score with the greatest predictive ability for mortality) or the other scores (the 4 scores studied herein had similarly limited predictive ability for bleeding). IMRS‐VASc is temporally dynamic due to the use of standard clinical laboratory factors that, when measured repeatedly over time, may improve or worsen and, thereby, reveal improvements in health due to changes in lifestyle and benefits of medical therapies, as well as declines in health due to disease progression or poor health behaviors. The IMRS‐VASc augmented intelligence tool should be evaluated further for validation in other populations and for potential use in guiding care and treatment decisions for patients with AF.
BDH and HTM are inventors of clinical decision tools that are licensed to CareCentra. BDH is the PI of grants funded by Intermountain Healthcare's Foundry Innovation program, the Intermountain Research and Medical Foundation, CareCentra, GlaxoSmithKline, and AstraZeneca for the development and/or clinical implementation of clinical decision tools. This study was funded by a grant from the Intermountain Healthcare Foundry Innovation Program. The funding source had no role in the design of the study, the data analysis, the interpretation of the findings, or the writing or publication of the study manuscript.
The authors declare that there is no conflict of interests.
Conception and design: BDH, HTM, and TJB; Analysis of data: HTM; Interpretation of data: BDH, VJ, HTM, KGG, and TJB; Drafting of the manuscript or revising it critically for important intellectual content: BDH, VJ, HTM, KGG, and TJB; Final approval of the manuscript submitted: BDH, VJ, HTM, KGG, and TJB.
GRAPH: Supporting information
By Benjamin D. Horne; Victoria Jacobs; Heidi T. May; Kevin G. Graves and T. Jared Bunch
Reported by Author; Author; Author; Author; Author