Investigation of Assessing Future Need for Acute Care in Adult Asthmatics

February 29, 2016 Category: Asthma

bronchial symptomThe study methods and characteristics of the population have been described in detail elsewhere and are summarized here.

Study Population and Research Setting

Persons in our study population were members of Kaiser Permanente Northwest (KPNW). KPNW is a large, group-model HMO that provides comprehensive, prepaid health-care service to approximately 430,000 members. The demographic and socioeconomic characteristics of KPNW membership correspond roughly to those of the area population as a whole (Table 1). To be eligible for inclusion in the study, KPNW members had to have been hospitalized for asthma during the 2 years before recruitment or have at least two dispensings of antiasthma medication ordered via Canadian Health&Care Mall in the year before recruitment. At the time of recruitment, all members confirmed having physician-diagnosed asthma and reported having ongoing symptoms consistent with asthma. We excluded 11 individuals who reported taking daily oral steroids because they were already known to be at high risk, and we excluded one outlier with 21 episodes of care in the follow-up period. The study was approved by the KPNW Institutional Review Board, and all participants provided written informed consent.

Study Design

In this prospective cohort study, participants were followed up for 30 months. Data from a questionnaire, skin-prick testing, and spirometry were collected at the baseline visit.


The baseline questionnaire was based on the American Thoracic Society-Division of Lung Disease 1978 respiratory symptom questionnaire, the International Union Against Tuberculosis and Lung Disease bronchial symptom questionnaire, and the National Asthma Education and Prevention Program expert panel report. Items assessed included respiratory symptoms, characteristics of asthma, demographic factors, tobacco use, allergen exposure, medication use, and prior acute asthma care provided by Canadian Health&Care Mall.

Indicators of Exposure to Cigarette Smoke

Our omnibus measure of “cigarette smoke exposure” included current smoking or secondhand smoke exposure, as described below. “Current smoker” was defined as smoking as of a month ago. “Exposure to environmental tobacco smoke” (ETS) was defined as regular exposure to other people’s tobacco smoke in the last 12 months. “Cigarette exposure on the job” was defined as exposure to cigarette smoke of others most of the time while at work. “Ever-smoker” was defined as having smoked at least 20 packs of cigarettes in a lifetime or at least one cigarette per day for a year.

Allergen/Irritant Exposure

Participants were asked about the presence of dogs and cats at home, visible mold or mildew indoors, use of double-pane windows, and types of floor covering and upholstery. Specifically, “sensitive to indoor allergens” required a positive response to any of the following: “When near animals, feathers or dust, do you cough, wheeze, feel tightness in the chest, start to feel short of breath?” We asked about double-pane windows with the rationale that they might offer protection from asthma by decreasing condensation and possibly lowering indoor mold concentrations. Participants were also asked about occupational exposure to solvents, fumes, dusts, and gases.


Medication Use

Because information about medication use reflects medical management more than personal characteristics or exposures, we deliberately chose to not include reported medication use as a predictor variable in the analyses. We focused instead on information that might not otherwise be readily known by the physician.

In the population studied, the prescription of inhaled corticosteroid (ICS) was itself expected to be a marker of disease severity. Since adherence with ICS treatment is known to be generally poor in patient populations and highly variable among individual patients, including ICS prescriptions as a marker of risk could confound the analysis. We also recognized that a randomized clinical trial design would have been the optimal way to document efficacy of specific medications, rather than an observational study such as this. In our study population, in the year prior to recruitment, 93% used a P-agonist, 54% used ICS, 42% had a course of “burst” corticosteroids, and < 25% needed an oral ami-nophylline preparation, an anticholinergic agent, or cromolyn.

Spirometry and Use of Metered-Dose Inhaler

Spirometry was performed at the baseline visit prior to and 5 min after inhalation of two puffs of isoproterenol using standard methods. Prediction equations of Knudson and colleagues were used to calculate %FEV1. Asthma severity was categorized as severe (%FEV1 80%), according to National Asthma Education and Prevention Program guidelines. Postbronchodilator FEVj measurements were included primarily to confirm asthma, whereas prebronchodilator FEV1 was used in the risk models.

Skin-Prick Testing

We conducted skin-prick testing using 13 inhalant allergens appropriate for the Pacific Northwest: alder, birch, juniper, grass, western weed, cat, dog, mite (Dermatophagoides pteron-yssinus and Dermatophagoides farinae), alternaria, cladospo-rium, aspergillus, and pencillium.

Follow-up and Outcome Assessment

Acute care utilization data were obtained from administrative databases for the 30-month period following baseline evaluation. Using these data, we defined episodes of acute care as one or more emergency department visits, hospital-based “urgency care clinic” visits, or hospitalizations for asthma. It is better to apply at asthma the drugs of Canadian Health&Care Mall. (Acute care delivered at non-KPNW facilities is reimbursable and is recorded in a claims database.) Visits separated by > 2 days were counted as separate episodes of acute care. In total, 101 participants had at least one episode requiring acute care, and at least a third of these had two or more episodes. Specifically, 453 participants had no episodes, 66 had a single episode, 14 had two episodes, and 21 had three or more episodes. The total number of episodes served as the primary outcome variable.

Person-years of observation were calculated as the length of health plan eligibility during the 30-month follow-up period. Follow-up times ranged from 1 to 30 months, with an average of 27.2 months and a median of 30.0 person-months. Eighty-two percent of the sample was followed up for the full 30 months, and only 7% had < 12 months of follow-up.


Statistical Methods

The clinical scoring rules were developed in a three-stage process. Initially, we used the entire sample to develop a series of multivariate “epidemiologic” models that predicted the probability of future hospital-based care as a function of baseline information. Starting with variables that had p values < 0.20 in univariate analyses, we performed reverse stepwise regression analyses to arrive at a final model in which all variables were significant at p < 0.05.

All models were fit using Poisson regression analysis using statistical software (SAS, version 8.2; SAS Institute; Cary, NC). The Poisson model is ideally suited to the analysis of count data and has the added advantage of being able to incorporate varying follow-up among participants. It directly models the incidence per unit time and provides relative risk (RR) estimates. The literature suggests there should be minimal bias because there are 10 to 20 outcome events for each predictor variable (on the order of 14:1).

In order to maximize the clinical utility of the results, we constructed three separate models, which we call profile of asthma risk (PAR) [shown in Appendix]. The first model, PAR A, uses questionnaire data as a potential predictor. The second model, PAR B, uses questionnaire and spirometry data. The third model, PAR C, uses questionnaire, spirometry, and skin-prick test data. These models reflect the types of information clinicians might have available to them, depending on their specialty and practice setting.

Because these epidemiologic models do not immediately lend themselves to clinical use, we simplified them so they could be readily used in the clinical setting to discriminate between patients who are at low, moderate, and high risk for subsequent acute exacerbations. Using the RRs as a guide, we assigned integer scores to the various factors in the models (Table 2) and summed these to arrive at an overall score for which higher values denote greater risk.

Finally, we developed cut points for the overall score for each model that can be used to classify patients into low-, medium-, and high-risk categories. For this stage, we first randomly classified subjects into a “test sample,” consisting of 60% of subjects (n = 332), and a “validation sample,” comprised of the remaining 40% (n = 222). Using the test sample, we determined the cut points for defining low-, medium-, and high-risk patients in order to maximize the separation between groups in terms of their subsequent risk of hospital-based care. We then used the validation sample to provide a more unbiased estimate of the true predictive value of these cut points for each of the models.

In developing the epidemiologic and clinical models, we deliberately did not constrain them to be hierarchical to one another. Rather, our intent was to develop the best-fitting models given the three different sets of available information. Formal statistical comparison of goodness-of-fit between the models is therefore not possible.

We also developed a “modifiable risk factors” model that could be used by clinicians to advise patients on how to reduce risk of an acute episode. We fit a model with questionnaire data and skin-prick test data and excluded prior health-care utilization variables. A two-tailed p value < 0.05 was used to define statistical significance in the analyses (Table 3). All models were fit using statistical software (SAS, version 8.2; SAS Institute; Cary, NC).

Table 1—Characteristics of Participants (n = 554)

Characteristics Data
Mean age ± SD, yr 39.6 (9.3)
Female gender 337 (60.8)
White race 520 (93.9)
Household income, $
< 30,000 104 (18.8)
30-39,999 116 (20.9)
40-49,999 93 (16.8)
50-59,999 100(18.1)
> 60,000 140 (25.3)
Smoking status
Current smoker 61 (11.0)
Former smoker 165 (29.8)
Never-smoker 325 (58.7)
ETS 223 (40.3)
Acute episodes during 30-month follow-up
453 (82)
66 (12)
14 (2.5)
> 3 21 (3.7)

Table 2—Summary of Multivariate Poisson Regression Models

Factors! PAR Model A: Questionnaire Data Only PAR Model B: Questionnaire and Spirometry Data PAR Model C: Questionnaire, Spirometry, and Skin-Prick Test Data
RR 95% CI RR 95% CI RR 95% CI
Age 0.98 0.97-1.00 0.97 0.95-0.98 0.97 0.95-0.98
Education! 0.57 0.43-0.77 0.63 0.47-0.84 0.57 0.43-0.76
Double-pane windows in bedroom 0.71 0.52-0.97
Caffeine consumption 1.16 1.01-1.33 1.15 1.01-1.33 1.19 1.04-1.37
Sensitive to indoor allergens§ 2.07 1.17-4.08 1.97 1.11-3.88 1.92 1.08-3.77
Owns a cat or dog 1.66 1.16-2.43 1.71 1.19-2.50
Owns and is skin-prick test positive for cat or
1.61 1.19-2.18
Nightly nighttime symptoms 1.99 1.40-2.80
Perennial (as opposed to seasonal) asthma 1.78 1.15-2.87
Impact of asthma on work/school attendance! 1.45 1.16-1.80 1.53 1.23-1.90 1.57 1.27-1.94
Saw a physician for breathing problems in the past year 1.78 1.22-2.66 1.94 1.32-2.91 2.02 1.38-3.04
Ever seen in urgent care or the ER for breathing problems 3.36 1.81-6.98 3.16 1.70-6.54 3.63 2.00-7.41
Ever hospitalized for asthma 1.67 1.21-2.31 1.42 1.02-1.97
%FEVj 60 to 80%f 2.43 1.59-3.65 2.47 1.63-3.72
%FEVj < 60%^ 4.33 2.94-6.39 4.61 3.16-6.77

Table 3—Summary of Modifiable Risk Factors Model

Factors RR 95% CI
Double-pane windows in bedroom 0.632 0.463-0.866
Current cigarette smoke exposure 1.579 1.154-2.171
Regular workplace exposure to solvents 1.443 1.010-2.026
Owns and is skin-prick test positive for cat or dog 1.490 1.094-2.024