To examine the effect of glucocorticoid use on the risk of various cardiovascular diseases in patients with polymyalgia rheumatica (PMR).
To examine the effect of glucocorticoid use on the risk of various cardiovascular diseases in patients with polymyalgia rheumatica (PMR).
We assembled a population-based incidence cohort of 364 patients with PMR first diagnosed between January 1, 1970 and December 31, 1999. Inclusion criteria were age ≥50 years, bilateral aching and morning stiffness involving at least 2 areas (neck, shoulders, hips, or proximal aspects of the thighs), and erythrocyte sedimentation rate (ESR) ≥40 mm/hour. In patients who fulfilled the first 2 criteria but had a normal ESR, a rapid response to low-dose glucocorticoids served as the third criterion. Patients were followed up until death or December 31, 2004. Cox models with time-dependent covariates were used to examine the association between glucocorticoid exposure and risk of myocardial infarction, heart failure, peripheral vascular disease, and cerebrovascular disease.
A total of 364 PMR patients (mean age 73 years, 67% women) were followed for a median of 7.6 years. During the disease course, 310 (85%) patients received glucocorticoids. After adjusting for age, calendar year, and ESR, patients who received glucocorticoids did not have a significantly higher risk for myocardial infarction, heart failure, peripheral vascular disease, or cerebrovascular disease (hazard ratio [95% confidence interval] 0.58 [0.29–1.18], 0.85 [0.45–1.54], 0.58 [0.24–1.40], and 0.65 [0.33–1.26], respectively) compared with those who did not receive glucocorticoids. In fact, a trend for a protective effect was seen. No significant association was observed between cumulative glucocorticoid dose and any of the outcomes (P = 0.39).
In patients with PMR, treatment with glucocorticoids is not associated with an increased risk of cardiovascular diseases.
The overall effect of glucocorticoids on the cardiovascular system and the risk of clinical cardiovascular events, especially among patients with inflammatory rheumatic diseases, remains unclear (1, 2). Glucocorticoids may indirectly increase the risk of cardiovascular disease through their adverse effects on traditional cardiovascular risk factors (3). Indeed, hypertension, hyperglycemia, dyslipidemia, and obesity are well-known adverse effects of glucocorticoids (3–5). Alternatively, glucocorticoids may have favorable effects on the cardiovascular system by alleviating systemic inflammation and reducing smooth muscle cell proliferation and ischemic tissue injury (6–9). There is clinical evidence supporting both directions of effect (10). In large database studies, increased risk has been observed among individuals taking high-dose glucocorticoids (11, 12), but the findings are inconsistent in rheumatic diseases, such as systemic lupus erythematosus or rheumatoid arthritis (2, 13–16). One potential explanation is that the increased risk of cardiovascular disease among persons taking glucocorticoids is, in fact, due to the underlying diseases for which they are prescribed (confounding by indication). Indeed, several of the underlying diseases themselves have been associated with a higher risk of cardiovascular disease (17, 18). Another potential explanation for the discrepant findings is that the adverse cardiovascular effects of glucocorticoids may differ according to disease or the disease state, such as at times of high versus low inflammatory activity (effect modification). One way to deal with these difficult questions is to study patients with different diseases and different levels of inflammatory activity.
Polymyalgia rheumatica (PMR) is an inflammatory rheumatic disease of the elderly characterized by aching and stiffness in the proximal regions of the extremities and torso and elevated markers of inflammation including erythrocyte sedimentation rate (ESR) (19). PMR is a common indication for long-term glucocorticoid therapy in the community. Although most patients are maintained on the lowest possible dose to control disease symptoms and to minimize glucocorticoid-induced complications, there is considerable heterogeneity in patients' clinical course and response to glucocorticoid therapy. Consequently, up to 50% of patients continue taking glucocorticoid therapy for ∼2–3 years or more (20). Therefore, patients with PMR constitute a unique population characterized by advanced age, high inflammatory activity, presence of several comorbid conditions, and long-term glucocorticoid therapy. We took advantage of these unique characteristics of patients with PMR to examine the effect of glucocorticoids on the risk of cardiovascular, peripheral vascular, and cerebrovascular diseases.
We conducted a retrospective, longitudinal, population-based study using the data resources of the Rochester Epidemiology Project (REP). The REP is a diagnostic indexing and medical records linkage system that provides ready access to the details of all medical care provided to local residents in Olmsted County, Minnesota (21, 22). Population-based studies are feasible in this community because it is geographically isolated, and most residents receive their medical care locally through a limited number of health care providers. Furthermore, all diagnoses and surgical procedures performed by these health care providers are indexed and retrievable, including diagnoses made during outpatient clinic consultations, emergency department visits, inpatient care, nursing home care, autopsy examination, or on death certificates. Consequently, the REP system assures virtually complete ascertainment of all clinically recognized conditions in a geographically defined community.
The study population comprised a previously described (20, 23) incidence cohort of all Olmsted County residents with PMR first diagnosed between January 1, 1970 and December 31, 1999. Briefly, individuals were included as patients with PMR if they fulfilled the following 3 criteria: age ≥50 years; bilateral aching and morning stiffness (lasting ≥30 minutes) persisting for at least 1 month involving 2 or 3 of the following areas: neck or torso, shoulders or proximal regions of the arms, and hips or proximal aspects of the thighs; and ESR increased to >40 mm/hour (Westergren). Patients with suggestive clinical findings (24) who fulfilled the first 2 of the 3 criteria and who had a prompt response to low-dose glucocorticoid therapy were also included. Except for giant cell arteritis (GCA), the presence of another disease that could explain the symptoms, such as active rheumatoid arthritis, was considered an exclusion criterion. In patients where the diagnosis was questionable, 3 rheumatologists reviewed all the clinical information and reached consensus on the diagnosis.
As described previously (20,23,25), data collection included details of PMR clinical and therapy characteristics recorded during all inpatient and outpatient encounters (including telephone calls) with the study patients for a period of 5 years from PMR incidence date. Clinical characteristics included disease relapses, recurrences, ESR values, and the occurrence of GCA based on American College of Rheumatology criteria (26). Details of nonsteroidal antiinflammatory drugs (NSAIDs) and glucocorticoid therapy were recorded at each encounter, including the route of administration and the doses. Glucocorticoid doses were recorded as prednisone equivalents.
Followup for PMR clinical and therapy characteristics started from the PMR incidence date and lasted until permanent remission of PMR (i.e., no PMR symptoms within 5 years of incidence date while patient was not receiving glucocorticoid therapy or was taking ≤5 mg/day of glucocorticoids and had normal ESR), migration from Olmsted County, or a maximum of 5 years after the incidence date. Followup for cardiovascular, peripheral vascular, and cerebrovascular events started prior to PMR incidence date (i.e., for events that occurred before PMR diagnosis) and lasted until migration from Olmsted County or December 31, 2004 (end of the study). Consequently, followup for these events was much longer than followup for PMR characteristics and extended beyond the 5-year duration of the disease. Four separate categories of outcomes were considered: myocardial infarction, heart failure, peripheral vascular events, and cerebrovascular events. Peripheral vascular events included aortic aneurysm, intermittent claudication, aortic dilatation, peripheral vascular disease, gangrene, peripheral vascular insufficiency, and vasospasm. Cerebrovascular events included subarachnoid hemorrhage; intracerebral hemorrhage; intracranial and subdural hemorrhage; occlusion and stenosis of extracranial carotid, vertebral, or basilar arteries; cerebral thrombosis; cerebral embolism; and transient ischemic attacks. All outcomes were ascertained passively using the electronic indices of diagnoses recorded from the complete community medical records.
Both oral and parenteral exposures to glucocorticoids were considered in this study. Glucocorticoid exposure was examined in 3 different ways. First, glucocorticoid exposure was categorized dichotomously according to whether or not glucocorticoids were used during the duration of the disease. Second, we examined the timing of glucocorticoid exposure in relation to the cardiovascular events using time-dependent covariates. Exposure was categorized as no use, recent use, or past use. Patients changed from one category to another during followup based on their glucocorticoid exposure. Recent use of glucocorticoids was defined as glucocorticoid use at the time of a cardiovascular event when a maximum of 1 year had elapsed since the last glucocorticoid dose. Past use was defined as discontinuation of glucocorticoids >1 year before the event. Third, we examined the cumulative glucocorticoid dose in tertiles of cumulative glucocorticoid dose categorized as low (≤35 mg), medium (>35 to ≤3,000 mg), and high (>3,000 mg).
Descriptive statistics (medians, proportions, etc.) were used to summarize the data. Cox regression models were used to estimate the influence of glucocorticoid exposure on the risk of each of the outcomes, adjusting for age and sex, calendar year, and ESR. Time-dependent covariates, which allow patients to change from one category of exposure to another during followup, were used to represent glucocorticoid exposure. Patients who experienced events prior to PMR diagnosis were excluded from each analysis, and patients were censored if GCA developed. Each outcome was examined separately, and a composite/combined outcome defined as the first of myocardial infarction, heart failure, peripheral vascular event, or cerebrovascular event during followup was also considered. In analyses of individual outcomes, we adjusted for the prior occurrence of any of the other cardiovascular outcomes. For example, if a patient experienced a myocardial infarction during the first year of PMR and subsequently developed heart failure, we adjusted for the prior occurrence of myocardial infarction in Cox models estimating the influence of glucocorticoid exposure on the risk of heart failure. We examined whether the risk estimates changed after adjusting for NSAID use in supplementary analyses, and after including patients with GCA in separate analyses. We also examined interactions between ESR values and timing of glucocorticoid exposure on the risk of cardiovascular, peripheral vascular, and cerebrovascular outcomes.
The study population comprised 364 patients with PMR who first fulfilled the diagnostic criteria for PMR between 1970 and 2000. The disease and therapy characteristics of the study population are outlined in Table 1. During the PMR disease course, 310 (85%) patients received glucocorticoids. Among the 310 glucocorticoid-treated patients, initial median glucocorticoid dose was 15 mg (interquartile range [IQR] 10–20) and the median duration of glucocorticoid therapy was 1.7 years (IQR 0.8–5.0). Average daily dose was 3.5 mg (IQR 1.0–5.0) and median cumulative dose was 3,158.7 mg (IQR 1,650.0–6,398.5) (Table 1). At the end of 5.1 years after PMR incidence, 67 (18%) patients were still receiving glucocorticoid therapy. Among the 67 patients still receiving glucocorticoid therapy, the median glucocorticoid dose was 5 mg (IQR 3–10).
|Female sex, no. (%)||244 (67.0)|
|Age at PMR diagnosis, years||73.4; 74.1 (67.5–79.4)|
|ESR at diagnosis, mm/hour||55.2; 52.0 (40.0–76.0)|
|Treatment characteristics, no. (%)|
|NSAIDs alone||54 (14.8)|
|Glucocorticoids alone||110 (30.2)|
|Glucocorticoids and NSAIDs||200 (55.0)|
|Glucocorticoid treatment characteristics (n = 310)|
|Initial dose, mg||19.2; 15.0 (10–20)|
|Duration of therapy, years||2.2; 1.7 (0.8–5.0)|
|Average daily dose, mg||3.5; 2.5 (1.0–5.0)|
|Cumulative dose, mg||4,803.5; 3,158.7 (1,650–6,398.5)|
Prior to PMR incidence, a total of 86 patients experienced at least 1 cardiovascular, cerebrovascular, or peripheral vascular event: 33 patients experienced myocardial infarction, 20 had heart failure, 22 experienced peripheral vascular events, and 36 experienced cerebrovascular events. Excluding these patients with a prior event, a total of 108 patients experienced at least 1 of the outcomes after PMR incidence: 47 experienced myocardial infarction, 68 experienced heart failure, 35 experienced peripheral vascular events, and 58 experienced cerebrovascular events. The cumulative incidence of all outcomes combined since time of PMR incidence is shown in Figure 1. By 5 years of followup, close to 30% of patients with PMR had at least 1 cardiovascular, cerebrovascular, or peripheral vascular event. The cumulative incidence reached 50% by 10 years. The cumulative incidence curve increased steadily throughout the 5 years of active disease course and thereafter, indicating that there was no evidence of a threshold. In other words, the risk for these outcomes remained roughly constant throughout the disease course and for several years thereafter.
The incidence rates of outcomes separately and combined among patients unexposed and exposed to glucocorticoids and the risk associated with glucocorticoid exposure after adjusting for age, sex, calendar year, and ESR are shown in Table 2. During 656 person-years of followup of patients not exposed to glucocorticoids, 31 patients experienced at least 1 of the outcomes for a rate of 8.4 per 100 person-years. During 2,492 person-years of followup of patients who were exposed to glucocorticoids, 77 patients experienced at least 1 of the outcomes for a rate of 5.6 per 100 person-years. Glucocorticoid exposure was associated with a 42% reduction in the risk of myocardial infarction (hazard ratio [HR] 0.58, 95% confidence interval [95% CI] 0.29–1.18) and peripheral vascular events (HR 0.58, 95% CI 0.24–1.40) and a 35% reduction in the risk of cerebrovascular events (HR 0.65, 95% CI 0.33–1.26). Risk reduction was less evident with heart failure (HR 0.85, 95% CI 0.45–1.54). When all events were combined, the risk estimate was marginally significant (HR 0.61, 95% CI 0.37–1.01).
|Outcome||Incidence in unexposed patients (no./person-years)†||Incidence in exposed patients (no./person-years)†||HR (95% CI)‡||P|
|Myocardial infarction||18/491.1||29/1819.0||0.58 (0.29–1.18)||0.580|
|Heart failure||19/491.3||49/1831.0||0.85 (0.45–1.54)||0.600|
|Peripheral vascular disease||9/560.6||26/1846.1||0.58 (0.24–1.40)||0.225|
|Cerebrovascular disease||17/490.9||41/1720.1||0.65 (0.33–1.26)||0.199|
|Combined outcome||31/369.3||77/1363.2||0.61 (0.37–1.01)||0.054|
The associations according to timing of glucocorticoid exposure in relation to occurrence of different outcomes are shown in Figure 2. For the combined end point, we observed a trend for a protective effect. Patients exposed to glucocorticoids in the past (at least 1 year prior to the event) had almost a 50% reduced risk (HR 0.52, 95% CI 0.29–0.94) compared with those who were never exposed. Recent exposure was also associated with a 31% reduced risk (HR 0.69, 95% CI 0.40–1.19), but the association did not reach statistical significance. In analyses according to the type of outcome, there was no noticeable difference between past versus recent exposure. The risk reduction for each of the outcomes was ∼20–50% except for heart failure, where the protective effect associated with glucocorticoid exposure appeared less strong for both recent and past use (Figure 2).
We then examined the association between cumulative glucocorticoid dose and the risk of outcomes, both separately and combined (Figure 3). There was no significant association between cumulative glucocorticoid dose and any cardiovascular, peripheral vascular, or cerebrovascular event (P = 0.39). Moreover, there was no indication for a dose-response relationship between the cumulative glucocorticoid dose and risk of either the individual events or combined outcome. Patients who received high cumulative doses of glucocorticoids (i.e., those who received >3,000 mg of prednisone equivalent during 5 years of followup) did not have a significantly higher risk for myocardial infarction (HR 0.59, 95% CI 0.27–1.27), heart failure (HR 0.81, 95% CI 0.41–1.60), peripheral vascular disease (HR 0.60, 95% CI 0.23–1.55), or cerebrovascular disease (HR 0.75, 95% CI 0.37–1.54) compared with those who did not receive glucocorticoids.
We performed additional analyses that included patients who were diagnosed with GCA at any time during followup. After inclusion of patients with GCA (instead of censoring), any glucocorticoid exposure was still associated with a reduced risk of the combined outcome (HR 0.68, 95% CI 0.43–1.06). Further adjustments for NSAID use also had no effect on the risk estimate (HR 0.64, 95% CI 0.39–1.05). There was no significant interaction between ESR values and timing of glucocorticoid exposure on the risk of these outcomes (P = 0.318), suggesting that the effect of glucocorticoid use on the risk of cardiovascular, cerebrovascular, or peripheral vascular events did not differ according to the level of inflammatory activity.
In this retrospective, population-based study, patients with PMR who received long-term glucocorticoid therapy for treatment of their PMR symptoms did not seem to have a higher risk of myocardial infarction, heart failure, peripheral vascular disease, or cerebrovascular disease. In fact, we observed a trend for a protective effect of glucocorticoids.
Our findings are in contrast with those of 2 large studies reporting an increased risk for cardiovascular disease among patients treated with glucocorticoids (11, 12). In a nested case–control study using the General Practice Research Database of the UK, glucocorticoid use was significantly more prevalent among patients with ischemic heart disease and heart failure compared with patients who did not experience these events (12). The association was stronger for current users, and among those who received the highest average daily dose. However, for patients with cerebrovascular events, glucocorticoids were associated with a protective effect. In another cohort study, glucocorticoid users and nonusers were followed up for ∼3 years through their administrative medical records for the occurrence of hospitalizations for myocardial infarction, revascularization procedures, heart failure, cerebrovascular events, and cardiovascular death (11). Patients receiving high daily doses of glucocorticoids (≥7.5 mg prednisolone equivalent) had almost a 3-fold increased risk of cardiovascular events. The association remained significant after adjustment for several cardiovascular risk factors and comedications and in subgroup analyses of various disease indications for glucocorticoid use.
There are a number of potential reasons why our findings differ from these 2 large studies demonstrating increased risk of cardiovascular disease among glucocorticoid users. First, both of these studies included >50,000 glucocorticoid users for a variety of disease indications. Consequently, the population of glucocorticoid users in these studies was very heterogenous in terms of their underlying diseases, level of disease activity and severity, and cardiovascular risk. Indeed, the cardiovascular effects in both studies differed in subgroups with different underlying diseases, such as rheumatoid arthritis, chronic obstructive pulmonary disease, or inflammatory bowel disease. For example, in one study (11), the glucocorticoid exposure was not significantly associated with cardiovascular events within the subgroup of patients with inflammatory bowel disease, indicating that the overall findings did not necessarily apply to all patient subgroups. Second, the glucocorticoid exposure definitions and the cardiovascular end points were different in each study and in ours, making it problematic to compare across studies. In one study (11), daily doses ≥7.5 mg prednisolone equivalent were considered as high doses and the low dose category included only inhaled, topical, and nasal glucocorticoids. In another study (12), high dose category included >20 mg prednisolone equivalent. In our study, we examined cumulative glucocorticoid dose instead of the average daily dose. Third, in these 2 studies, patients with a history of cardiovascular events prior to the initiation of glucocorticoids were not excluded, further complicating the interpretation of findings. Both studies attempted to account for confounding by indication and disease activity by adjusting for underlying indications and comedications. However, residual confounding is a potential issue given the very heterogenous study populations and inability to account for inflammatory disease activity. Our study population, albeit much smaller, included a much more homogenous population of elderly patients with repeated ESR measurements to assess inflammatory activity and detailed glucocorticoid exposure data. Taken together, these findings suggest that the adverse cardiovascular effects of glucocorticoids may differ according to the disease state, and in patients with PMR, the overall effect on the risk of clinical cardiovascular events is potentially favorable.
In contrast to the increased cardiovascular risk observed among patients receiving high doses of glucocorticoids in large administrative database studies, findings are inconsistent in patients with rheumatic diseases such as systemic lupus erythematosus or rheumatoid arthritis (2, 13–15). Interestingly, recent reviews challenge the widely held opinion that prolonged administration of glucocorticoids accelerates atherosclerosis and increases the risk for cardiovascular events (13, 16, 27). For example, patients with lupus without atherosclerotic plaques received significantly higher daily doses of glucocorticoids than did those with plaques, suggesting that more aggressive control of inflammatory disease activity in systemic lupus erythematosus might help prevent atherosclerosis (28). In patients with rheumatoid arthritis, glucocorticoid use was associated with carotid plaque and arterial incompressibility, independent of cardiovascular risk factors and rheumatoid arthritis clinical manifestations (15). In another rheumatoid arthritis study that examined clinical events, early glucocorticoid use (in the first 3 years of disease) was associated with a higher risk of cardiovascular events, whereas cumulative glucocorticoid dose more than 1 year before cardiovascular events was associated with a lower risk, especially in the subset of patients who experienced cardiovascular events (29, 30). In a series of 47 patients with rheumatoid arthritis who were treated long term with a very high mean cumulative dose of prednisone (15.9 grams), no correlation between the cumulative corticosteroid dose and the presence of severe subclinical atherosclerotic findings (31) or left ventricular diastolic dysfunction (32) was found. Similarly, no correlation between the cumulative dose of prednisone (mean 15.7 grams) and the development of endothelial dysfunction was observed in 55 patients with longstanding rheumatoid arthritis without cardiovascular disease (33). Data from our rheumatoid arthritis group indicated that the risk associated with high-dose glucocorticoid use was in fact confined mainly to patients with positive rheumatoid factor and differed according to presence of history of ischemic heart disease (17, 34). So far, there are no published studies that examined the association between glucocorticoid use and cardiovascular events in patients with PMR.
Given that glucocorticoids profoundly inhibit systemic inflammatory and immune mediators and modulate the vascular endothelium and local inflammatory processes (8, 35, 36), it is conceivable that glucocorticoid therapy attenuates the cardiovascular risk among the subset of patients with high inflammatory activity. The effects of glucocorticoids on the complex network of biochemical processes leading to cardiovascular events may depend upon the genetic background, underlying disease pathology, metabolic and inflammatory status of the patient, and the extent of existing atherosclerosis. These effects may vary for different cardiovascular events. An individual's immune/inflammatory response is only one part of many biochemical processes leading to cardiovascular events, and glucocorticoids may alter only one part of this inflammatory response. In other words, glucocorticoids possibly affect only one part of the overall orchestration of these complex processes, and the effects clearly may vary depending on the underlying disease status.
This study extends the findings of previous studies by being the first to examine cardiovascular effects of glucocorticoids in elderly patients with PMR who continue glucocorticoid therapy over an extended period. This was a population-based inception cohort sampled from a well-defined community and closely monitored for the occurrence of study end points. We collected detailed encounter-specific data on inflammatory disease activity (ESR), NSAIDs, and glucocorticoid therapy during the PMR disease course and used time-dependent variables for both ESR values and glucocorticoid exposure to capture changes in inflammatory disease activity and glucocorticoid exposure levels throughout followup. The PMR cohort had a median of 44.4 years of medical history prior to their PMR incidence and 7.6 years following PMR, allowing us to record observations (e.g., cardiovascular history) made several years before the PMR incidence date and continuing for several years following therapy. Therefore, selection bias and information bias were minimized in our study. Altogether, these unique strengths allowed us to elucidate the potential contribution of glucocorticoid therapy to the risk of cardiovascular events in this unique inflammatory disease setting.
Our findings must be interpreted in light of a number of potential study limitations. We were not able to account for various potential confounders, especially cardiovascular risk factors and concomitant medical conditions. Yet, most of these potential confounders would have confounded the association towards the null and are unlikely to explain the protective effect observed in our study. All of the analyses were adjusted for calendar year and we did not observe any significant cohort effects. The cardiovascular outcomes were based on physician diagnosis and were not validated using epidemiologic diagnostic criteria, such as the Framingham heart failure criteria (37). However, the comprehensive medical records system of the REP allowed us to access several years of complete medical history, which is not feasible in any other setting. In this geographically well-defined population, virtually all residents come to medical attention in a 3-year period, therefore differential ascertainment bias is less likely to affect our results. Channeling bias is a potential limitation if the treating physicians were less likely to prescribe glucocorticoids to patients who had a preexisting cardiovascular risk. We excluded patients with preexisting cardiovascular events from the analysis. We also investigated whether the prior occurrence of any of the cardiovascular outcomes during followup was associated with a change in risk for the other outcomes and found no significant relationship. Given that the standard treatment of PMR is with glucocorticoids, almost all of the patients were exposed to glucocorticoids. The study sample was primarily white, limiting the generalizability of our findings to more ethnically diverse populations. Finally, the statistical power was limited as demonstrated by the wide confidence intervals. Given that the confidence intervals include values as high as 2.5, we cannot exclude the possibility of an increased risk.
In conclusion, our findings indicate that glucocorticoid treatment in PMR is not associated with an increased risk for myocardial infarction, heart failure, peripheral vascular events, or cerebrovascular events. Our observation is consistent with previous reports of decreased relative mortality in patients with PMR (25, 38). Given the conflicting findings of cardiovascular effects of glucocorticoids in different studies, we conclude that the beneficial and adverse cardiovascular effects of glucocorticoids may differ according to disease and disease state.
Dr. Gabriel 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.
Study design. Maradit Kremers, Crowson, Davis, Hunder, Gabriel.
Acquisition of data. Reinalda, Crowson.
Analysis and interpretation of data. Maradit Kremers, Crowson, Davis, Hunder, Gabriel.
Manuscript preparation. Maradit Kremers, Crowson, Davis, Hunder, Gabriel.
Statistical analysis. Reinalda, Crowson.
The authors wish to acknowledge Margaret Donohue, RN, for performing data abstraction.