[Skip to Content]
Sign In
Individual Sign In
Create an Account
Institutional Sign In
OpenAthens Shibboleth
[Skip to Content Landing]
Figure 1.  Algorithm Development and Validation Strategy
Algorithm Development and Validation Strategy

aEligible visit days were defined as days when blood pressure (BP) measurements occurred on the same day as eligible services. Visits occurring on a day that included nursing home or hospital care were excluded.

IHC indicates intensive hypertension treatment; and VA, Department of Veterans Affair.

Figure 2.  Final BP Medication Identification Algorithm
Final BP Medication Identification Algorithm
Table 1.  Characteristics of Visits Included in the Analytic Sample
Characteristics of Visits Included in the Analytic Sample
Table 2.  Acute and Chronic Conditions Contributing to IHC
Acute and Chronic Conditions Contributing to IHC
Table 3.  Top 7 Chronic Condition Categories Associated With False-Positive Algorithm IHC and Contributing Conditions to IHC
Top 7 Chronic Condition Categories Associated With False-Positive Algorithm IHC and Contributing Conditions to IHC
1.
Staessen  JA, Gasowski  J, Wang  JG,  et al.  Risks of untreated and treated isolated systolic hypertension in the elderly: meta-analysis of outcome trials.   Lancet. 2000;355(9207):865-872. doi:10.1016/S0140-6736(99)07330-4PubMedGoogle ScholarCrossref
2.
Wright  JT  Jr, Williamson  JD, Whelton  PK,  et al; SPRINT Research Group.  A randomized trial of intensive versus standard blood-pressure control.   N Engl J Med. 2015;373(22):2103-2116. doi:10.1056/NEJMoa1511939PubMedGoogle ScholarCrossref
3.
Williamson  JD, Supiano  MA, Applegate  WB,  et al; SPRINT Research Group.  Intensive vs standard blood pressure control and cardiovascular disease outcomes in adults aged ≥75 years: a randomized clinical trial.   JAMA. 2016;315(24):2673-2682. doi:10.1001/jama.2016.7050PubMedGoogle ScholarCrossref
4.
Gueyffier  F, Bulpitt  C, Boissel  JP,  et al; INDANA Group.  Antihypertensive drugs in very old people: a subgroup meta-analysis of randomised controlled trials.   Lancet. 1999;353(9155):793-796. doi:10.1016/S0140-6736(98)08127-6PubMedGoogle ScholarCrossref
5.
Beckett  NS, Peters  R, Fletcher  AE,  et al; HYVET Study Group.  Treatment of hypertension in patients 80 years of age or older.   N Engl J Med. 2008;358(18):1887-1898. doi:10.1056/NEJMoa0801369PubMedGoogle ScholarCrossref
6.
Rogers  MA, Ward  K, Gure  TR,  et al.  Blood pressure trajectories prior to death in patients with diabetes.   Diabetes Care. 2011;34(7):1534-1539. doi:10.2337/dc11-0441PubMedGoogle ScholarCrossref
7.
Ravindrarajah  R, Hazra  NC, Hamada  S,  et al.  Systolic blood pressure trajectory, frailty, and all-cause mortality >80 years of age: cohort study using electronic health records.   Circulation. 2017;135(24):2357-2368. doi:10.1161/CIRCULATIONAHA.116.026687PubMedGoogle ScholarCrossref
8.
Odden  MC, Peralta  CA, Haan  MN, Covinsky  KE.  Rethinking the association of high blood pressure with mortality in elderly adults: the impact of frailty.   Arch Intern Med. 2012;172(15):1162-1168. doi:10.1001/archinternmed.2012.2555PubMedGoogle ScholarCrossref
9.
Hayward  RA.  All-or-nothing treatment targets make bad performance measures.   Am J Manag Care. 2007;13(3):126-128.PubMedGoogle Scholar
10.
Kerr  EA, Smith  DM, Hogan  MM,  et al.  Comparing clinical automated, medical record, and hybrid data sources for diabetes quality measures.   Jt Comm J Qual Improv. 2002;28(10):555-565. doi:10.1016/S1070-3241(02)28059-1PubMedGoogle Scholar
11.
Kerr  EA, Zikmund-Fisher  BJ, Klamerus  ML, Subramanian  U, Hogan  MM, Hofer  TP.  The role of clinical uncertainty in treatment decisions for diabetic patients with uncontrolled blood pressure.   Ann Intern Med. 2008;148(10):717-727. doi:10.7326/0003-4819-148-10-200805200-00004PubMedGoogle ScholarCrossref
12.
Kerr  EA, Lucatorto  MA, Holleman  R, Hogan  MM, Klamerus  ML, Hofer  TP; VA Diabetes Quality Enhancement Research Initiative (QUERI) Workgroup on Clinical Action Measures.  Monitoring performance for blood pressure management among patients with diabetes mellitus: too much of a good thing?   Arch Intern Med. 2012;172(12):938-945. doi:10.1001/archinternmed.2012.2253PubMedGoogle ScholarCrossref
13.
World Health Organization.  International Classification of Diseases, Ninth Revision (ICD-9). World Health Organization; 1977.
14.
Steinman  MA, Lee  SJ, John Boscardin  W,  et al.  Patterns of multimorbidity in elderly veterans.   J Am Geriatr Soc. 2012;60(10):1872-1880. doi:10.1111/j.1532-5415.2012.04158.xPubMedGoogle ScholarCrossref
15.
Sussman  JB, Kerr  EA, Saini  SD,  et al.  Rates of deintensification of blood pressure and glycemic medication treatment based on levels of control and life expectancy in older patients with diabetes mellitus.   JAMA Intern Med. 2015;175(12):1942-1949. doi:10.1001/jamainternmed.2015.5110PubMedGoogle ScholarCrossref
16.
Anderson  TS, Jing  B, Auerbach  A,  et al.  Clinical outcomes after intensifying antihypertensive medication regimens among older adults at hospital discharge.   JAMA Intern Med. 2019. doi:10.1001/jamainternmed.2019.3007PubMedGoogle Scholar
17.
Song  W, Intrator  O, Lee  S, Boockvar  K.  Antihypertensive drug deintensification and recurrent falls in long-term care.   Health Serv Res. 2018;53(6):4066-4086. doi:10.1111/1475-6773.13074PubMedGoogle ScholarCrossref
18.
Mehta  PA, Cowie  MR.  Gender and heart failure: a population perspective.   Heart. 2006;92(suppl 3):iii14-iii18. doi:10.1136/hrt.2005.070342PubMedGoogle ScholarCrossref
Limit 200 characters
Limit 25 characters
Conflicts of Interest Disclosure

Identify all potential conflicts of interest that might be relevant to your comment.

Conflicts of interest comprise financial interests, activities, and relationships within the past 3 years including but not limited to employment, affiliation, grants or funding, consultancies, honoraria or payment, speaker's bureaus, stock ownership or options, expert testimony, royalties, donation of medical equipment, or patents planned, pending, or issued.

Err on the side of full disclosure.

If you have no conflicts of interest, check "No potential conflicts of interest" in the box below. The information will be posted with your response.

Not all submitted comments are published. Please see our commenting policy for details.

Limit 140 characters
Limit 3600 characters or approximately 600 words
    Views 601
    Citations 0
    Original Investigation
    Geriatrics
    July 30, 2020

    Validation of a Health System Measure to Capture Intensive Medication Treatment of Hypertension in the Veterans Health Administration

    Author Affiliations
    • 1VA Ann Arbor Medical Center, Geriatric Research, Education, and Clinical Center, Ann Arbor, Michigan
    • 2Division of Geriatric and Palliative Medicine, Department of Internal Medicine, University of Michigan, Ann Arbor
    • 3VA Center for Clinical Management Research, Health Services Research and Development Center of Innovation, Ann Arbor, Michigan
    • 4Institute of Healthcare Policy and Innovation, University of Michigan, Ann Arbor
    • 5Division of General Internal Medicine, Department of Internal Medicine, University of Michigan, Ann Arbor
    • 6Institute for Social Research, University of Michigan, Ann Arbor
    • 7Preventive Medicine Section, Memphis VA Medical Center, Memphis, Tennessee
    • 8Department of Preventive Medicine, University of Tennessee Health Science Center, Memphis
    • 9Department of Medicine, Section of Geriatrics, Yale School of Medicine, New Haven, Connecticut
    • 10Consulting for Statistics, Computing and Analytics Research, University of Michigan, Ann Arbor
    • 11Department of Biostatistics, University of Michigan Medical School, Ann Arbor
    • 12Accelerate Discovery and Practice Transformation Center of Innovation, VA Health Care System, Durham, North Carolina
    • 13Department of Population Health Sciences, Duke University, Durham, North Carolina
    JAMA Netw Open. 2020;3(7):e205417. doi:10.1001/jamanetworkopen.2020.5417
    Key Points español 中文 (chinese)

    Question  Can health system data be used to accurately detect patients receiving intensive hypertension care with multiple blood pressure medications?

    Findings  This cross-sectional study used data from 319 patients with 3625 visits to develop an algorithm based on clinically measured blood pressure and pharmacy fills to detect intensive hypertension care delivered at any visit. Using electronic health record review of clinical notes as the reference standard to detect patients aged 65 years or older receiving 3 or more medications with systolic blood pressure of less than 120 mm Hg, the algorithm had a sensitivity of 92% and a specificity of 97%.

    Meaning  The findings of this study suggest that this algorithm could provide a resource for health care systems to measure high-intensity care for quality comparison or as a research tool.

    Abstract

    Importance  Blood pressure (BP) targets are the main measure of high-quality hypertension care in health systems. However, BP alone does not reflect intensity of pharmacological treatment, which should be carefully managed in older patients.

    Objectives  To develop and validate an electronic health record (EHR) data-only algorithm using pharmacy and BP data to capture intensive hypertension care (IHC), defined as 3 or more BP medications and BP less than 120 mm Hg, and to identify conditions associated with greater IHC, either through greater algorithm false-positive IHC, or by contributing clinically to delivering more IHC.

    Design, Setting, and Participants  This cross-sectional study was conducted among 319 randomly selected patients aged 65 years or older receiving IHC from the Veterans Health Administration (VHA) from July 1, 2011, to June 30, 2013. Data were collected from a total of 3625 primary care visits. Data were analyzed from January 2017 to March 2020.

    Exposures  Calibration and measurement of the algorithm for IHC (algorithm IHC).

    Main Outcomes and Measures  For each primary care visit, the reference standard, clinical IHC, was determined by detailed review of free-text clinical notes. The correlation in BP medication count between the EHR-only algorithm vs the reference standard and the sensitivity and specificity of the algorithm IHC were calculated. In addition, presence vs absence of contributing conditions acting in combination with hypertension management were measured to examine incidence of IHC associated with contributing conditions, including an acute condition that lowered BP (eg, dehydration), another condition requiring a BP target lower than the standard 140 mm Hg (eg, diabetes), or the patient needing a BP-lowering medication for a nonhypertension condition (eg, β-blocker for atrial fibrillation) resulting in low BP.

    Results  Among 319 patients with 3625 visits (mean [SD] age, 75.6 [7.2] years; 3592 [99.1%] men), 911 visits (25.1%) had clinical IHC by the reference standard. The algorithm for determining medication count was highly correlated with the reference standard (r = 0.84). Sensitivity of detecting clinical IHC was 92.2% (95% CI, 89.3%-95.1%), and specificity was 97.2% (95% CI, 96.1%-98.3%), suggesting that clinical IHC can be identified from routinely collected data. Only 75 visits (2.1%) were algorithm IHC false positives, 55 visits (1.5%) involved IHC with contributing conditions, and 125 visits (3.5%) involved either false-positive or IHC with contributing conditions. Among select contributing conditions, congestive heart failure (37 patients [5.2%]) was most associated with a prespecified combined false-positive or IHC with contributing conditions rate higher than 5%.

    Conclusions and Relevance  These findings suggest that health system data can be used reliably to estimate IHC.

    Introduction

    Lower blood pressure (BP) and the intensive treatment with BP medications required to achieve it prevent strokes and death.1-5 However, if achieving low BP requires multiple medicines, this could increase adverse effects and treatment burden. The current system of performance measures is designed to increase BP control, but it does not discourage intensive treatment that far exceeds the target (eg, by >20 mm Hg). In addition, BP declines naturally before death,6,7 and BP control may be less beneficial among adults who are frail.8 Therefore, further research is needed to support more clinically nuanced care for older adults who might not need such intensive hypertension care (IHC). However, before such research can be conducted in health systems, the first step is to determine whether IHC using multiple BP medications can be identified using health system data.

    We have previously described approaches to assess BP level and medication treatment simultaneously from extractable electronic health record (EHR) data fields to improve large-scale health system efforts to achieve appropriate BP treatment.9-12 However, there are reasons the EHR measure could be misleading about the actual care provided. A patient may fill a medication but later discontinue it owing to adverse effects, therefore using fewer medications than the pharmacy data would suggest.

    We conducted this cross-sectional study to develop and validate the accuracy of an electronic measure of IHC for older patients in the Veterans Health Administration (VHA). In addition, we sought to understand which comorbid conditions, in combination with hypertension treatment, could potentially contribute synergistically to the delivery of IHC (eg, by increasing measurement error or because their treatment also requires BP medication treatment). The results could allow patients with those conditions to be analyzed separately or excluded from future quality improvement initiatives or health outcomes research.

    Methods

    This cross-sectional observational study was approved by the Department of Veterans Affairs (VA) institutional review board and Centers for Medicare & Medicaid Service data were provided by VA Health Services Research and Development Service, VA Information Resource Center. A waiver of informed consent was granted because the VHA dataset is large, and obtaining informed consent was considered not feasible for this study, per VA policy. This study is reported following the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.

    Data

    We used 2009 through 2013 national VA Clinical Data Warehouse data, including all outpatient clinical encounters, vital signs, diagnoses, pharmacy records, and a file of non-VHA medications. We also linked 2011 to 2013 Medicare Part D drug claims to Clinical Data Warehouse data.

    Sample

    We identified a cohort of older patients (ie, age ≥65 years) with hypertension receiving VHA primary care during a 2-year period from July 1, 2011, to June 30, 2013. Eligibility was determined as having at least 1 hypertension (International Classification of Diseases, Ninth Revision [ICD-9] code 401.x13) diagnosis and an established primary care relationship (ie, ≥2 primary care visits in 2009-2011 with ≥1 of those visits in 2009 to 2010).

    The unit of analysis was each eligible visit with physicians, nurses, advanced practice clinicians, pharmacists, or social workers in primary care (ie, family medicine, general or internal medicine, geriatrics, or mental health primary care), or specialty care clinicians who manage hypertension (ie, nephrologists, endocrinologists, cardiologists, and neurology outpatient nonemergency care clinicians) (eTable 1 in the Supplement). We defined eligible visits as days when BP measurements occurred on the same day as eligible services. We excluded visits occurring on a day that included nursing home or hospital care. eTable 2 in the Supplement provides code for constructing the measures.

    To enhance our sample with a greater opportunity to study variations in IHC than in a general VHA sample, we used a preliminary algorithm (ie, prior to calibration) of IHC to identify patients with a continuous period of IHC12: systolic BP less than 120 mm Hg on at least 2 consecutive primary care visits, with both visits preceded by VHA BP medication fills of at least 3 different classes (any dose) within 100 days. Out of 1 260 150 older VHA patients with hypertension, the 60 899 veterans (4.8%) meeting these criteria had a mean (SD) of 12.5 (8.8) visits during the study period. Of 13 923 patients who entered the study period with prevalent IHC, only 1617 patients (11.6%) continued to have uninterrupted IHC for the rest of the study period. The overall mixture of visits was balanced between 79 704 episodes of contiguous IHC and 55 539 isolated IHC visits. From this sampling frame, we randomly selected 420 patients for further study (Figure 1).

    Overview of Study Design

    We first developed an algorithm to measure IHC using EHR extractable data fields only as the test. Then we reviewed the text-based clinical notes for documented BP medications, BP, and the clinical care provided (hereafter, the reference standard). Last, we measured the accuracy of the IHC EHR data algorithm against the reference standard.

    Algorithm Development Using Pharmacy and BP EHR Data Only

    To calibrate the medication algorithm, we structured the EHR data as a longitudinal series of BP medication fill events (from VHA data only) before and after each outpatient visit (Figure 2). Medications that reduced BP were organized into classes: (1) angiotensin agents as a single class containing both angiotensin-converting enzyme inhibitors (ACEIs) and angiotensin receptor blockers, (2) calcium channel blockers, (3) thiazide, (4) potassium-sparing diuretics, (5) β-blockers, (6) centrally acting α-agonists, (7) other vasodilators, and (8) direct renin blockers (eTable 3 in the Supplement). Fills of different medications within a class would be considered a switch between the medications rather than an additional medication.

    At each visit, we considered whether or not a patient was receiving a BP medication class using the pharmacy fill data prior to (ie, the prefill) and after each visit (ie, the refill). First, we considered continuously refilled BP medications. A prefill and a refill in the same class (Figure 2), regardless of the dose, was considered a match. For example, if patient had a lisinopril prefill, then a subsequent losartan fill (representing a switch within class) would be considered as the matched refill. We calibrated the algorithm on a subsample of 30 patients who had 455 visits and 1749 medication fill events. We first varied the criteria for days between the prefill and the eligible visit (ie, the look-back period) between 120 and 365 days to assess changes in false positives (positive algorithm test but the reference standard determined nonuse) and false negatives (negative algorithm test but the reference standard documented use). Second, we similarly varied the refill criteria. We allowed for time greater than the typical 90-day supply to bridge over a refill at a non-VHA pharmacy or verbal instructions to decrease the dose (eg, pill-splitting, which would increase the effective supply).

    Next, we developed criteria for discontinued medications. If we found a medication with a prefill but no refill, we considered the medication to be potentially discontinued (ie, not active on the visit day). For each medication, we calculated time elapsed since the prefill date divided by number of supplied days, resulting in a percentage of days’ supply elapsed by the visit day (range, 0% to ≥100%). For example, if the prefill was for a 90-day supply and the visit was 30 days since the fill, then 33% of supply days were elapsed. We then calibrated the most accurate threshold in percentage of days’ supply elapsed. We calibrated these thresholds separately depending on whether the potentially discontinued medication was chronic (defined as prefill with a prior fill within the past year) or new. Compared with continued medications, fewer medications were discontinued, so we used the entire analytic sample of 3625 visits to calibrate the discontinuation algorithm.

    After algorithm calibration, we considered pharmacy data from 2 additional sources. First, we considered the non-VHA medication file maintained voluntarily by VHA clinicians. Despite reducing false-negative rate by 7.5%, this data source worsened false positives by 45.0%. Therefore, we dropped this data source from further analysis. Second, 67 patients (21.0%) also used Medicare Part D. The proportion of visits with at least 3 medications increased from 63.8% in the VHA-only pharmacy data to 67.8% after adding Medicare D data, so we retained Part D data for subsequent analysis.

    EHR Review

    Trained EHR abstractors (including L.G. and A.L.) reviewed free-text clinical notes to obtain key elements for comparison with EHR and pharmacy data. First, abstractors added documented BP data on eligible visits not already captured in the EHR data (eg, if manually rechecked during the visit or from a home BP log on the same day as the visit). Next, to determine the daily BP medications, reviewers read all eligible visit notes as well as interval notes between eligible visits (eg, telephone notes, emergency visits, hospital discharge summaries), including scanned non-VHA notes.

    Reference Standard Review of Visits for Contributing Conditions

    We reviewed the clinical notes for 3 types of contributing conditions, circumstances in which a condition may have been associated with IHC at the visit (ie, acting in combination synergistically with hypertension treatment). The 3 IHC with contributing conditions situations were (1) an acute condition spuriously lowering BP (eg, dehydration), (2) another chronic condition requiring a BP target lower than 140 mm Hg (eg, diabetes), and (3) a need to take a BP-lowering medication for a nonhypertension chronic condition (eg, a β-blocker for heart rate control in atrial fibrillation) that was documented as causing low BP. More than 1 contributing condition could be coded for any IHC with contributing conditions visit. A total of 10% of EHRs were abstracted by 2 coders (L.G. and A.L.) for interrater reliability in presence vs absence of each class of BP medications and any contributing conditions vs no contributing conditions (pooled κ = 0.92).

    Statistical Analysis

    We used manual review of clinical notes as the reference standard and the medication algorithm as the test to count the number of medications at each visit. If more than 1 systolic BP was measured in a visit, we calculated the systolic BP as the mean to reflect underlying true BP. If the clinical notes documented additional systolic BPs (eg, rechecked by the clinician or noted on a home BP log), then it was possible for the reference standard to differ from the EHR-only systolic BP.

    We calculated (1) agreement as κ between the EHR and reference standard for the presence vs absence of each BP medication class and (2) the correlation in total number of medications at each visit. Next, we classified each visit as representing IHC or not (defined as ≥3 BP medication classes and mean systolic BP <120 mm Hg) by the reference standard and by the final developed algorithm. We calculated sensitivity, specificity, positive predictive value, and negative predictive value of the algorithm IHC. All calculations adjusted SEs by clustering of visits within patient. Next, we added the false-positive rate with the IHC with contributing conditions rate to estimate the percentage of visits that the algorithm IHC potentially overestimated hypertension care intensity. We calculated this summary measure in the overall sample and by chronic condition subsamples (ie, diabetes, coronary heart disease, cardiovascular disease, arrhythmia, heart failure [HF], chronic kidney disease, psychiatric conditions, and benign prostatic hypertrophy). We set a 5% rate as the clinically significant proportion of false-positives (either due to a data error or because a comorbidity was the reason for IHC). We hypothesized that IHC would be more likely to be overestimated (owing to false-positive and IHC with contributing conditions) for conditions also treated with BP-lowering medications (eg, HF). The chronic conditions were determined by any coded visit diagnoses during the observation window, and all visits for that patient were considered associated with those conditions.

    Data were analyzed using SAS version 9.4 (SAS Institute) and Stata version 14 (StataCorp) statistical software. Data analysis was conducted from January 2017 to March 2020.

    Results

    Of 420 randomly sampled patients, we used 31 EHRs to train EHR abstractors. We discarded these EHRs. Of 389 remaining patients, we excluded 58 patients who died during the observation period, 7 who spent 10% or more of the study time in inpatient or nursing home care, 3 patients enrolled in a hypertension treatment trial, and 2 patients for missing Medicare pharmacy files (Figure 1). The analytic sample of 3625 visits for 319 remaining patients (mean [SD], 11.4 [8.3] visits per patient) included 3582 visits (99.1%) from men, with a mean (SD) age of 75.6 (7.2) years. The prevalence of other chronic conditions was comparable to the general older VHA population,14 including 1090 visits (30.1%) from patients with diabetes, 706 visits (19.5%) from patients with congestive HF, and 1042 visits (28.7%) from patients with coronary artery disease (Table 1). The reference standard mean (SD) count of BP medications was 2.58 (0.96). The mean (SD) systolic BP was 121.8 (16.4) mm Hg. Among 3625 total visits, 911 visits (25.1%) met criteria for IHC by the reference standard review (Table 1).

    Algorithm Development to Determine Active Medication Use on Each Visit Day

    For calibration of continuous refills in the subset of 30 patients with 455 visits, we found a steady decrease in false negatives when broadening the lookback window from 14.2% at 120 days to 10.8% at 365 days. The false-positive rate was low and varied little when the look-back window changed, starting at 1.2% at 120 days, increasing to 1.3% at 136 days, but increasing to 1.3% between 180 and 200 days. We set our prefill and refill at 186 days (31 days × 6 months).

    On the full sample, 3625 visits were associated with 10 594 medications with valid prefills within 186 days. After 9017 prefilled medications (85.1%) were matched with a valid refill, we calibrated the remaining prefilled medications without a valid refill (1577 medications [14.9%]) as potentially discontinued medications (eFigure in the Supplement). The false-positive and false-negative rates were optimized if the visit date occurred before 80% of the days’ supply elapsed for discontinuation of new medications and before 90% of the day’s supply elapsed for old medications. The discontinuation criteria restored 511 true-positive medications (4.8%), compared with an algorithm that required medications to be refilled to be counted. These analyses resulted in the final algorithm (Figure 2).

    Validating Algorithm to Identify IHC

    In the analytic sample of 319 patients, the count of BP medications in the algorithm (mean [SD], 2.59 [0.98]; range 0-6) vs the reference standard (mean [SD], 2.58 [0.96]; range 0-6) was highly correlated (r = 0.84 [95% CI, 0.79-0.89]). The correlation was also high for the algorithm applied to VHA-only (ie, without Medicare Part D) pharmacy data (r = 0.84 [95% CI, 0.83-0.85]). The most common medications were ACEIs or angiotensin receptor blockers (2870 visits [79.2%]) and β-blockers (2899 visits [80.0%]). The agreement for each class ranged from κ = 0.83 (β-blocker and ACEI or angiotensin receptor blockers) to κ = 0.96 (potassium-sparing diuretics).

    We found positive algorithm IHC at 915 visits (25.3%) vs the reference standard at 911 visits (25.1%). The sensitivity was 92.2% (95% CI, 89.3%-95.1%), specificity was 97.2% (95% CI, 96.1%-98.3%), positive predictive value was 91.8% (95% CI, 88.5%-95.1%), and negative predictive value was 97.4% (95% CI, 96.4%-98.4%), suggesting that IHC could be validly identified from routinely collected data. Of 3625 visits, 75 visits (2.1%) were false positives due to errors in EHR data.

    Conditions Contributing to IHC

    A positive algorithm IHC visit could also have a contributing condition. Of 3625 visits, the reference standard review identified 174 visits (4.8%) with a potential contributing condition, but only 55 visits (1.5%) were also algorithm IHC positives (Table 2), including 16 visits (0.4%) with an acute contributing condition, 16 visits (0.4%) with an intentionally low BP target owing to a contributing condition, and 24 visits (0.7%) with a treatment tradeoff due to treatment of the contributing condition.

    There were 125 visits (3.5%) that were potentially overestimated owing to false positive or IHC with contributing condition (Table 3). Of comorbid conditions, only 37 patients with congestive HF (5.2%) had a total potential combined rate of either false positive or IHC with contributing conditions that exceeded 5%. The next highest conditions with overestimated IHC were prostate disease (13 patients [5.0%]) and psychiatric condition (14 patients [4.1%]).

    Discussion

    In this cross-sectional study using large national health care system data, we developed a method to capture IHC, defined as 3 or more BP medications with systolic BP less than 120 mm Hg. Our calibrated algorithm for IHC had sensitivity of 92.2% and specificity of 97.2%, with an overall low total rate of false positives (2.1% of all visits).

    We were also able to test whether certain comorbid conditions contributed to false-positive IHC or situations in which a nonhypertension condition contributed or mixed effects with hypertension treatment to result in clinical IHC. Patients with congestive HF were the most likely to have visits of either type of overestimation, thus future applications of IHC may need to stratify by patient condition or remove patients with congestive HF. We may have underestimated the rate of less prevalent conditions in IHC with contributing conditions owing to small sample size. The condition with the second highest prevalence of overestimation, benign prostatic hypertrophy, may be specific to the VHA, owing to wide use of terazosin in the VHA serving as a double-duty drug for hypertension and urinary retention. The third ranked condition, psychiatric condition, may have been due to dual use of clonidine for posttraumatic stress disorder.

    This study builds on methods previously developed to use health system data to measure the adequacy of clinical action, not only by measuring the percentage of patients with BP below a certain threshold, but also capturing the intensity of BP medications prescribed.12 By integrating BP values and medications, we gain a more comprehensive view of the care provided to patients by a clinician or a health system than by using BP alone. Using only BP as the target (eg, BP <140/90 mm Hg, the current performance measure in the VHA and by health insurances) is a far simpler computation. By capturing medication intensity, it is possible to capture more clinical nuance: from intense treatment (low BP and multiple medications), to mild undertreatment (moderate hypertension on treatment), to severe undertreatment (no medications despite high BP). This method paves the way beyond dichotomous BP targets for future research based on intensity of care.

    Medication data combined with BP yields greater understanding of systems-based management than BP alone. We have found previously that intensively treated BP for patients with diabetes is rarely deintensified, with little effect by advanced age or mortality risk.15 Research by others has captured IHC accurately by computerized administration records in more acute settings: intensifying hypertension regimens on hospital discharge worsens clinical outcomes of older adults,16 and intensely treated patients in postacute nursing homes with history of falls are less likely to fall again after deintensification.17 This study now broadens our understanding of administrative data accuracy for capturing IHC in the outpatient setting. This method will be important in coming years after the Systolic Blood Pressure Intervention Trial (SPRINT)2 has redefined what we consider to be beneficial yet high-intensity treatment. How we defined IHC (ie, BP <120 mm Hg and using ≥3 BP medications) is consistent with the SPRINT intervention, in which the goal was to attain a target BP of less than 120 mm Hg by increasing medications prescribed. We note here that the period of time covered by these analyses preceded publication of the SPRINT results, during a time when the systolic BP goal within the VHA was less than 140 mm Hg. These patients may have been treated intensively above and beyond the VHA goals at the time, but in the post-SPRINT era, this degree of intensity may be more commonly attained for many older patients.

    Strengths and Limitations

    This study has some strengths. The main strength was its national patient representation and its use of multiple sources of data. The algorithm was calibrated using only VHA pharmacy data. The inclusion of Medicare Part D data added to the complete picture; however, those data were not critical to the accuracy of the algorithm. Therefore, our results would be applicable to health systems or insurance plan populations for whom most pharmacy data are obtainable. Second, our method calibrated the algorithm to capture IHC at any patient visit, therefore allowing flexibility for future outcomes studies to study the combination of visits needed to construct measures of short- or long-term exposure to IHC. For example, one could study patients with year-over-year change in IHC (by comparing last visit IHC of each year) or select patients with consecutive visits of IHC to identify those with consistent long-term IHC.

    There are several limitations that must be acknowledged. First, we had to ensure that our sample was intensively treated and had variability between visits to have an opportunity to test for the absence or presence of IHC. Therefore, we studied patients with generally intensive BP care, possibly resulting in a false-positive rate that is higher than in the rest of the VHA. Second, for a portion of our algorithm (ie, medications that were discontinued), we needed to use the entire data set for calibration then validated back using the same sample, therefore potentially overestimating the algorithm accuracy. Third, we were unable to distinguish patients with HF with reduced ejection fraction, which we expect would be more likely than preserved ejection fraction to contribute to finding IHC with contributing conditions, because HF with reduced ejection fraction is appropriately treated with BP-lowering medications to achieve BP below usual hypertension goals. Fourth, our algorithm could not capture free samples of BP medications dispensed by non-VHA clinics and BP medications purchased at lower cost than VHA copayments (eg, $5 generic programs). Last, this older population was predominantly male; therefore, we could not test for sex differences in algorithm accuracy. An older cohort of women might be treated less aggressively for comorbid conditions,18 potentially limiting our IHC with contributing conditions analysis among women.

    Conclusions

    This cross-sectional study’s focus on medication counts is the first step in validating a high-intensity BP regimen, but there is still room for further research regarding dosing intensity. Current guidelines for comorbid conditions now recommend treatment with BP-lowering medications, such as low-dose ACEIs for renal protection in diabetes. Therefore, capturing total doses in addition to medications prescribed will reflect IHC with greater granularity in the context of guideline-consistent care.

    We have developed and validated an algorithm to identify patients with IHC in health systems. This method could be used to compare IHC among clinicians or systems or to improve appropriateness of patient care.

    Back to top
    Article Information

    Accepted for Publication: March 17, 2020.

    Published: July 30, 2020. doi:10.1001/jamanetworkopen.2020.5417

    Open Access: This is an open access article distributed under the terms of the CC-BY License. © 2020 Min L et al. JAMA Network Open.

    Corresponding Author: Lillian Min, MD, MSHS, Division of Geriatric and Palliative Medicine, Department of Internal Medicine, University of Michigan, 300 N Ingalls Bldg, Floor 9, Ann Arbor, MI 48109 (lmin@umich.edu).

    Author Contributions: Dr Min had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

    Concept and design: Min, Hofer, Cushman, Kim, Maciejewski.

    Acquisition, analysis, or interpretation of data: Min, Ha, Hofer, Sussman, Langa, Tinetti, Kim, Gillon, Larkin, Chan, Kerr.

    Drafting of the manuscript: Min, Ha, Hofer.

    Critical revision of the manuscript for important intellectual content: Min, Sussman, Langa, Cushman, Tinetti, Kim, Maciejewski, Gillon, Larkin, Chan, Kerr.

    Statistical analysis: Min, Ha, Sussman, Kim, Maciejewski, Chan.

    Obtained funding: Min, Kim, Kerr.

    Administrative, technical, or material support: Sussman, Maciejewski, Gillon, Larkin, Chan.

    Supervision: Hofer.

    Conflict of Interest Disclosures: Dr Hofer reported receiving grants from the Department of Veteran Affairs (VA) Health Services Research and Development Service (HSR&D) and National Heart, Lung, and Blood Institute during the conduct of the study. Dr Sussman reported receiving grants from the VA during the conduct of the study. Dr Langa reported grants from the National Institutes of Health (NIH) during the conduct of the study. Dr Cushman reported receiving grants from the NIH and NIA during the conduct of the study; and grants from Eli Lilly, personal fees from Sanofi, and consulting for Novartis outside the submitted work. Dr Maciejewski reported receiving grants from the VA HSR&D during the conduct of the study and owing stock in Amgen owing to his spouse's employment. Dr Kerr reported receiving grants from the VA HSR&D and NIA during the conduct of the study and personal fees from Bind Insurance outside the submitted work. No other disclosures were reported.

    Funding/Support: This material is based on work supported by grant No. IIR 14-083 from the Department of Veterans Affairs, Veterans Health Administration, Office of Research and Development, Health Services Research and Development (VA HSRD) and grant No. R01 AG047178 from the National Institute on Aging. Dr Maciejewski received support from grant No. RCS 10-391 from the VA HSRD.

    Role of the Funder/Sponsor: The funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

    References
    1.
    Staessen  JA, Gasowski  J, Wang  JG,  et al.  Risks of untreated and treated isolated systolic hypertension in the elderly: meta-analysis of outcome trials.   Lancet. 2000;355(9207):865-872. doi:10.1016/S0140-6736(99)07330-4PubMedGoogle ScholarCrossref
    2.
    Wright  JT  Jr, Williamson  JD, Whelton  PK,  et al; SPRINT Research Group.  A randomized trial of intensive versus standard blood-pressure control.   N Engl J Med. 2015;373(22):2103-2116. doi:10.1056/NEJMoa1511939PubMedGoogle ScholarCrossref
    3.
    Williamson  JD, Supiano  MA, Applegate  WB,  et al; SPRINT Research Group.  Intensive vs standard blood pressure control and cardiovascular disease outcomes in adults aged ≥75 years: a randomized clinical trial.   JAMA. 2016;315(24):2673-2682. doi:10.1001/jama.2016.7050PubMedGoogle ScholarCrossref
    4.
    Gueyffier  F, Bulpitt  C, Boissel  JP,  et al; INDANA Group.  Antihypertensive drugs in very old people: a subgroup meta-analysis of randomised controlled trials.   Lancet. 1999;353(9155):793-796. doi:10.1016/S0140-6736(98)08127-6PubMedGoogle ScholarCrossref
    5.
    Beckett  NS, Peters  R, Fletcher  AE,  et al; HYVET Study Group.  Treatment of hypertension in patients 80 years of age or older.   N Engl J Med. 2008;358(18):1887-1898. doi:10.1056/NEJMoa0801369PubMedGoogle ScholarCrossref
    6.
    Rogers  MA, Ward  K, Gure  TR,  et al.  Blood pressure trajectories prior to death in patients with diabetes.   Diabetes Care. 2011;34(7):1534-1539. doi:10.2337/dc11-0441PubMedGoogle ScholarCrossref
    7.
    Ravindrarajah  R, Hazra  NC, Hamada  S,  et al.  Systolic blood pressure trajectory, frailty, and all-cause mortality >80 years of age: cohort study using electronic health records.   Circulation. 2017;135(24):2357-2368. doi:10.1161/CIRCULATIONAHA.116.026687PubMedGoogle ScholarCrossref
    8.
    Odden  MC, Peralta  CA, Haan  MN, Covinsky  KE.  Rethinking the association of high blood pressure with mortality in elderly adults: the impact of frailty.   Arch Intern Med. 2012;172(15):1162-1168. doi:10.1001/archinternmed.2012.2555PubMedGoogle ScholarCrossref
    9.
    Hayward  RA.  All-or-nothing treatment targets make bad performance measures.   Am J Manag Care. 2007;13(3):126-128.PubMedGoogle Scholar
    10.
    Kerr  EA, Smith  DM, Hogan  MM,  et al.  Comparing clinical automated, medical record, and hybrid data sources for diabetes quality measures.   Jt Comm J Qual Improv. 2002;28(10):555-565. doi:10.1016/S1070-3241(02)28059-1PubMedGoogle Scholar
    11.
    Kerr  EA, Zikmund-Fisher  BJ, Klamerus  ML, Subramanian  U, Hogan  MM, Hofer  TP.  The role of clinical uncertainty in treatment decisions for diabetic patients with uncontrolled blood pressure.   Ann Intern Med. 2008;148(10):717-727. doi:10.7326/0003-4819-148-10-200805200-00004PubMedGoogle ScholarCrossref
    12.
    Kerr  EA, Lucatorto  MA, Holleman  R, Hogan  MM, Klamerus  ML, Hofer  TP; VA Diabetes Quality Enhancement Research Initiative (QUERI) Workgroup on Clinical Action Measures.  Monitoring performance for blood pressure management among patients with diabetes mellitus: too much of a good thing?   Arch Intern Med. 2012;172(12):938-945. doi:10.1001/archinternmed.2012.2253PubMedGoogle ScholarCrossref
    13.
    World Health Organization.  International Classification of Diseases, Ninth Revision (ICD-9). World Health Organization; 1977.
    14.
    Steinman  MA, Lee  SJ, John Boscardin  W,  et al.  Patterns of multimorbidity in elderly veterans.   J Am Geriatr Soc. 2012;60(10):1872-1880. doi:10.1111/j.1532-5415.2012.04158.xPubMedGoogle ScholarCrossref
    15.
    Sussman  JB, Kerr  EA, Saini  SD,  et al.  Rates of deintensification of blood pressure and glycemic medication treatment based on levels of control and life expectancy in older patients with diabetes mellitus.   JAMA Intern Med. 2015;175(12):1942-1949. doi:10.1001/jamainternmed.2015.5110PubMedGoogle ScholarCrossref
    16.
    Anderson  TS, Jing  B, Auerbach  A,  et al.  Clinical outcomes after intensifying antihypertensive medication regimens among older adults at hospital discharge.   JAMA Intern Med. 2019. doi:10.1001/jamainternmed.2019.3007PubMedGoogle Scholar
    17.
    Song  W, Intrator  O, Lee  S, Boockvar  K.  Antihypertensive drug deintensification and recurrent falls in long-term care.   Health Serv Res. 2018;53(6):4066-4086. doi:10.1111/1475-6773.13074PubMedGoogle ScholarCrossref
    18.
    Mehta  PA, Cowie  MR.  Gender and heart failure: a population perspective.   Heart. 2006;92(suppl 3):iii14-iii18. doi:10.1136/hrt.2005.070342PubMedGoogle ScholarCrossref
    ×