Subject recruitment and retention

We used data collected during a prospective, longitudinal cohort study of adults with asthma recruited from physician practices in Northern California. Details of the study design have been previously reported [23,24,25,26]. Each subject underwent a structured, computer-assisted telephone interview covering demographic characteristics, smoking history, asthma history, symptoms, and treatment, health status, health care utilization for asthma, and insurance for asthma care.

Physicians registered 669 eligible patients. After initial data collection at baseline (n = 601) and 18-month follow-up interviews (n = 539), we later restricted the data set to 371 of the baseline cohort (55% of total registry) and 242 of the follow-up subjects (65% of restricted baseline cohort). We restricted the data set to eliminate all interviews potentially compromised by faulty data collection or documentation by a single survey interviewer [26,27]. This restricted data set excluded 24 baseline subjects who were found to be outside the study age range and 206 baseline subjects with inconsistent data during subsequent re-interview. Of the 371 baseline subjects, the present study excludes an additional 129 subjects at 18-month follow-up interview who had inconsistent data at later interviews or did not complete follow-up, leaving 242 follow-up interviews (18 month). These exclusions had no significant effect on study findings.

Demographic data for comparison of the baseline cohort (n = 371) with registered subjects (n = 669) are not available. Compared with subjects who participated in both baseline and follow-up interviews (n = 242), subjects without complete follow-up interviews (n = 129) were younger (36.6 years versus 40.5 years) and less likely to have white race/ethnicity (62% versus 71%; P < 0.001 and P = 0.10, respectively). There were no statistical differences in history of ever smoking (43% of participants in both interviews versus 37% of non-participants at follow-up), female gender (73% versus 66%), atopic history (82% versus 83%), or severity-of-asthma scores (11.0 versus 10.6; P > 0.15 in all cases).

Hospitalization for asthma

The primary study outcome was self-reported hospitalization for asthma during the 18-month prospective follow-up period. Subjects were asked at 18-month follow-up interviews whether they had been hospitalized for asthma during the previous 18 months. Although subjects could indicate more than one positive response, we analyzed the binary outcome of one or more hospitalization for asthma.

Risk factor variables

All demographic variables were based on baseline subject interview responses. Current and prior cigarette smoking history was assessed using questions adapted from the National Health Interview Survey [28].

We previously developed and validated a 13-item disease-specific severity-of-asthma score with four subscales: frequency of current asthma symptoms (daytime or nocturnal), use of systemic corticosteroids, use of other asthma medications (besides systemic corticosteroids), and history of hospitalizations and intubations [23–25]. Possible total scores range from 0 to 28, with higher scores reflecting more severe asthma. To examine the relative impact of recent and remote hospitalization on further hospitalization for asthma over longitudinal follow-up, we removed hospitalization from the established severity score and defined two new variables: recent hospitalization (during the 12 months prior to baseline interview), or remote hospitalization (past hospitalization not meeting the previous definition of recent). As a result, the hospitalization and intubation subscale now reflects only prior history of intubation.

Several other clinical aspects of asthma were assessed. We defined asthma onset as the subject-reported age of first asthma symptoms. Atopic history was defined by a reported history of allergic rhinitis or atopic dermatitis. Because prior work suggests an unexpectedly high prevalence of aspirin intolerance in persons with near fatal asthma [29], we ascertained any history of aspirin sensitivity at baseline interview. Since gastroesophageal reflux disease (GERD) may exacerbate asthma symptoms [30], we evaluated whether subjects were taking H2-blockers or proton pump inhibitors as surrogates for GERD and related conditions. We furthermore determined whether subjects possessed a peak flow meter for home usage.

Generic health status was measured using the Medical Outcomes Study SF-36 questionnaire [31,32]. We assessed several indicators of health care access for asthma care, including whether subjects had a regular site for asthma care, a principal care provider for asthma, medical insurance for outpatient asthma care, and an annual deduction for outpatient medical care. We also identified subjects who appeared to rely on emergency department (ED) services for urgent asthma care. We defined reliance on ED care as one or more self-reported ED visits during the interview interval but no urgent outpatient clinic or office visits for asthma, either to regular or alternate sources of asthma care.

Statistical analysis

In a previous analysis of pulmonary and allergy specialist care, the severity-of-asthma score was associated with an increased risk of hospitalization at 18-month follow-up [25]. The current study evaluates other risk factors for asthma-related hospitalization, taking baseline asthma severity into account. Because asthma severity may act on the causal pathway between a risk factor and subsequent hospitalization for asthma, we present multivariate models both including and excluding baseline asthma severity and generic health status. For example, low education could increase the risk of hospitalization either directly, through poor self-management strategies, or indirectly, if poorly educated persons have greater asthma severity for other reasons. We also delineate the components of asthma severity — respiratory symptoms, systemic corticosteroid use, other asthma medication use, past intubations, and previous hospitalizations — that are most strongly predictive of subsequent hospitalization.

Interview data were analyzed using SAS 6.12 software (SAS Institute, Cary, NC, USA). We evaluated the association between baseline characteristics and the risk of hospitalization for asthma during the ensuing 18-month follow-up period, reported at the 18-month interview. We use the data set restricted to 242 subjects with verified baseline and follow-up interviews for all analyses.

Bivariate relationships were examined using logistic regression analysis, with separate models for each predictor variable. We used multiple logistic regression analysis to elucidate the independent association between each baseline variable and the prospective risk of hospitalization. In constructing the multivariate model, all predictor variables whose bivariate odds ratio and 95% confidence interval suggested a possible association with hospitalization were entered into the final model. All variables deemed important on an a priori basis, such as age, were also included.

Health care access

Reflecting the sampling method employed, all subjects identified a regular source of asthma care and a primary medical provider for asthma care at baseline interview. The majority of participants (97%) also reported having health insurance covering outpatient visits for asthma. Approximately one-third of subjects indicated having annual insurance deductible for physician visits (31%). The majority of subjects continued to report health insurance coverage (96%) and ongoing primary asthma care (99%) at 18-month follow-up. Despite apparent access to outpatient medical care, a substantial proportion (16%) appeared to rely on the ED for urgent asthma care.

Demographic factors and the risk of hospitalization: bivariate analysis

Thirty-nine subjects (16%) reported at least one hospitalization for asthma during the prospective 18-month follow-up period. Of the baseline characteristics, non-white race (OR, 2.1; 95% CI, 1.1–4.0) and lower income (OR, 1.3; 95% CI, 1.1–1.5) were associated with a greater risk of hospitalization for asthma during the 18-month follow-up (Table 1). Current smokers had an increased likelihood of hospitalization, although the confidence interval did not exclude no association. Greater educational attainment was related to a lower risk of hospitalization (OR, 0.8 per year of education; 95% CI, 0.70–0.96).

Clinical risk factors for hospitalization: bivariate analysis

Better baseline generic physical health status (SF-36) was associated with a slightly decreased risk of subsequent hospitalization (6% reduction in odds per 1 point score increment; 95% CI, 3-9%) (Table 2). Mental health status was furthermore associated with a 4% reduction in the risk of hospitalization per 1 point (95% CI, 0–7%). Reliance on emergency department care was finally related to a greater risk of hospitalization (OR, 2.5; 95% CI, 1.1–5.5).

Risk of hospitalization — multivariate analysis

We examined the relation between race and hospitalization in more detail. African–American race was associated with an increased risk of hospitalization for asthma, compared with white, non-Hispanic persons, after controlling for covariates (OR, 10.2; 95% CI, 1.8–58.4). Hispanic race/ethnicity also appeared related to hospitalization (OR, 4.0; 95% CI, 0.9–18.0). There was no apparent relation between Asian race and risk of hospital admission (OR, 2.0; 95% CI, 0.4–10.9).

To further examine the association between asthma severity and hospitalization, we repeated the multivariate analysis dividing the overall severity-of-asthma score into its four subscales. The systemic corticosteroid score (OR, 1.7 per 5-point score increment; 95% CI, 1.3–2.3) and recent hospitalization for asthma (OR, 9.7; 95% CI, 2.2–43.0) were significantly associated with an increased risk of asthma hospitalization, after controlling for covariates. There was, conversely, no statistical relationship between other asthma medications (OR, 1.1; 95% CI, 0.8–1.6) or asthma symptom scores (OR, 1.2; 95% CI, 0.7–1.9) and the ensuing risk of hospitalization. Systemic corticosteroid use and recent asthma hospitalization, then, appear to drive the relationship between asthma severity and hospitalization for asthma.

Because asthma severity could act as a causal intermediate between a risk factor and the risk of hospitalization, we repeated the multivariate analysis excluding asthma severity and generic health status from the model (Table 3). In this analysis, low income was more strongly related to a greater risk of hospitalization for asthma (OR, 1.2; 95% CI, 1.02–1.4). Use of gastric acid suppression therapy was also associated with increased risk (OR, 2.2; 95% CI, 1.0–4.9). Although other point estimates and confidence intervals changed slightly, there were no other notable changes compared with the model controlling for asthma severity.

To examine whether subjects without baseline health insurance coverage (3%) were affecting study results, we repeated the multivariate analysis excluding these subjects. Only one of the 39 subjects hospitalized at follow-up had no baseline health insurance. There was no meaningful impact on the results in all multivariate analyses. For example, the estimate for lower income in the model without asthma severity was nearly unchanged (OR, 1.2; 95% CI, 1.03–1.4).

Subject recruitment and retention

In brief, we recruited a random sample of certified American Board of Medical Specialty pulmonary specialists and allergy/immunology specialists. The participating physicians maintained a registry of persons aged 18–50 years with outpatient visits for asthma over a prospective 4-week period. Each person registered by a participating physician was contacted to arrange a telephone interview.

Risk factor variables

We ascertained household income as a series of increments (less than $5000, $5001–$10,000, $10,000 increments up to $50,000, $50,001–$75,000, and above $75,000). To convert to specific income levels, the mid-increment value was applied (except for the highest category, where we used a value of $87,500).

The severity-of-asthma score systemic corticosteroid sub-scale includes items for any past corticosteroid use, use during the past 12 months, and steroid dependency. The asthma medication subscale incorporates current use of inhaled corticosteroids, inhaled non-steroidal anti-inflammatory agents, inhaled bronchodilators, oral beta agonists and theophylline-containing medications, nasal medications (antihistamines, decongestants, and topical corticosteroids), and home nebulizer use. In this subscale, one point is assigned for each medication used during the past 2 weeks, with one additional point for frequent inhaled beta agonist or corticosteroid use.

We measured generic health status using the Medical Outcomes Study SF-36 questionnaire. Previous work demonstrates the SF-36 instrument's validity in adult asthma [31]. The physical and mental component summary scores have been defined from the eight SF-36 subscales by factor analysis [32], which measure physical and mental dimensions of health, respectively. Each summary score in the general US population has a mean of 50 and a standard deviation of 10. Higher scores reflect more favorable health states.