Trajectories of Adherence to Airway Clearance Therapy forPatients with Cystic Fibrosis

Avani C. Modi,1 PHD, Amy E. Cassedy,2 PHD, Alexandra L. Quittner,3 PHD, Frank Accurso,4 MD,

Marci Sontag,5 PHD, Joni M. Koenig,6 MSPH, and Richard F. Ittenbach,7 PHD1Cincinnati Children’s Hospital Medical Center, University of Cincinnati College of Medicine, 2Cincinnati

Children’s Hospital Medical Center, 3University of Miami, 4University of Colorado Denver, 5Colorado School

of Public Health, University of Colorado Denver, 6The Children’s Hospital of Denver, and 7Cincinnati

Children’s Hospital Medical Center, University of Cincinnati College of Medicine

All correspondence concerning this article should be addressed to Avani C. Modi, PhD, Division of

Behavioral Medicine and Clinical Psychology, Center for the Promotion of Adherence and Self Management,

Cincinnati Children’s Hospital Medical Center, 3333 Burnet Avenue – MLC 7039, Cincinnati, OH 45229,

USA. E-mail: avani.modi@cchmc.org

Received November 25, 2009; revisions received January 19, 2010; accepted February 15, 2010

Objective Although cross-sectional studies have demonstrated poor adherence to airway clearance therapy

(ACT) for patients with cystic fibrosis (CF), no studies have identified longitudinal patterns of adherence.

The objective was to characterize and identify predictors of ACT adherence trajectories for individuals

with CF. Methods Secondary data analyses were conducted for a randomized clinical trial examining

differences in three ACTs. Participants (n¼ 153; M¼ 14.3 years, 55% male, 86% Caucasian, baseline FEV1%

predicted: M¼ 86.7)/primary caregivers completed Daily Phone Diaries, an empirically supported adherence

measure, every 4 months. Results Group-based trajectory modeling revealed the best-fitting solution was

a three-group model: low-adherence (14%), medium-adherence (49%), and high-adherence (37%) groups.

ACT type was the only significant predictor of adherence trajectories. Discussion Three trajectories of

adherence to ACT for patients with CF were found. With the identification of trajectories, adherence inter-

ventions can be targeted for the subgroup at highest risk in order to prevent poor health outcomes.

Key words chest wall oscillation; compliance; developmental; Flutter; group-based models; high frequency;longitudinal; manual chest physiotherapy.

Cystic fibrosis (CF) is a progressive, multi-system pulmon-

ary disease, affecting approximately 30,000 individuals in

the United States (Cystic Fibrosis Foundation, 2009).

Significant advances in medicine have resulted in increased

lifespan for individuals with CF, shifting from 25 years of

age in 1985 to 37.4 years in 2008 (Cystic Fibrosis

Foundation, 2009). Treatment for CF has become increas-

ing complex in the past two decades, often taking patients

several hours to complete each day. Treatment often in-

volves taking several types of pills, including oral antibiot-

ics, pancreatic enzymes, digestive medications, and

vitamins. Because lung disease is the primary cause of

morbidity and mortality in patients with CF, much of the

treatment regimen is devoted to thinning and reducing

mucus, as well as keeping the airways open. This is accom-

plished by prescribing nebulized medications (e.g., bron-

chodilators, dornase alpha for mucus thinning, and

inhaled antibiotics) and airway clearance therapy (ACT).

Evidence from several clinical trials and a recent review

examining the benefits of ACT have identified improved

mucus transport as one of the most prominent benefits

of this treatment (see Bradley, Moran, & Elborn, 2006

for evidence-based review). Several forms of ACT are cur-

rently prescribed, including postural drainage and percus-

sion (PD&P), Flutter, and high-frequency chest wall

oscillation (HFCWO), also known as Vest therapy.

Journal of Pediatric Psychology 35(9) pp. 1028–1037, 2010

doi:10.1093/jpepsy/jsq015

Advance Access publication March 18, 2010

Journal of Pediatric Psychology vol. 35 no. 9 � The Author 2010. Published by Oxford University Press on behalf of the Society of Pediatric Psychology.All rights reserved. For permissions, please e-mail: journals.permissions@oxfordjournals.org

at University of M

iami - O

tto G. R

ichter Library on February 13, 2014

http://jpepsy.oxfordjournals.org/D

ownloaded from

Although ACT remains a mainstay of CF treatment at

this time, it is time-consuming and can be one of the most

difficult to incorporate into daily routines, often taking as

much as an hour a day (Abbott, Dodd, & Webb, 1996; Modi

& Quittner, 2006a). Not surprisingly, several cross-

sectional studies have suggested poor adherence to ACT

across childhood (Modi et al., 2006; Passero, Remor, &

Salomon, 1981; White, Miller, Smith, & McMahon,

2009), adolescence (Quittner et al., 2000; White et al.,

2009), and adulthood (Abbott et al., 1996; Conway,

Pond, Hamnett, & Watson, 1996; White, Stiller, &

Haensel, 2007). For example, school-aged children demon-

strated ACT adherence rates of 51–74% (Modi et al., 2006).

ACT adherence has also been shown to be poor during ado-

lescence (50%; Quittner et al., 2000) and adulthood

(30–32%; Abbott et al., 1996; Myers & Horn, 2006), with

increasing age associated with worse adherence

(Arias Llorente, Bousono Garcia, & Diaz Martin, 2008). In

addition to age, pulmonary functioning has also been asso-

ciated with adherence in CF (Zindani, Streetman,

Streetman, & Nasr, 2006). In contrast, studies are lacking

or there are inconclusive results regarding the association

between adherence and sex (Myers, 2009; Patterson, Wall,

Berge, & Milla, 2008) and socioeconomic status in CF.

However, it is possible that similar to other pediatric condi-

tions (Berquist et al., 2006; Modi, Morita, & Glauser, 2008),

these may be critical factors affecting ACT adherence.

Although these studies provide important snapshots

of adherence at various developmental stages, no studies

have examined adherence to ACT over time for patients at

different points in their lifespan. Identifying the long-term

patterns of adherence in patients with chronic disease is

important because it allows us to understand their variabil-

ity and potentially identify critical periods for intervention.

Furthermore, it may be the first step in identifying thresh-

olds for efficacy related to treatment. For example, what

level of adherence is necessary to obtain optimal health

outcomes, 75, 80, or 100%? This is especially true for

patients with CF, whose treatment regimen only becomes

more complex over time as a result of worsening symptoms

and deterioration of health. While longitudinal ACT adher-

ence data are unavailable, recent adherence data for neb-

ulized medications using an adaptive aerosol delivery

device indicated significant intra- and inter-patient variabil-

ity in adherence (Latchford, Duff, Quinn, Conway, &

Conner, 2009; McNamara, McCormack, McDonald,

Heaf, & Southern, 2009), which was then used to person-

alize and subsequently improve adherence to this treat-

ment (e.g., reduce treatment from twice to once daily;

(McNamara et al., 2009). Overall, these studies highlight

the need to examine longitudinal patterns of adherence for

individuals with CF, allowing us to target individuals at

highest risk during critical developmental periods.

Empirically supported treatments for adherence are

behaviorally and family based, often requiring eight to 12

sessions (Kahana, Drotar, & Frazier, 2008). However,

these interventions are not readily available (e.g., limited)

to the populations that need them most. By identifying

which patients are at highest risk, the limited resources

needed to improve adherence through intervention can

be better targeted across child and adult chronic disease

populations. Targeting patients at highest risk may also

improve health outcomes, such as pulmonary functioning,

reduction in hospitalizations for pulmonary exacerbations,

the likelihood of becoming infected with drug-resistant

bacteria, and health-related quality of life. In addition,

determining the patient-specific (e.g., age, sex) and

treatment-specific (e.g., ACT type) factors that impact

adherence over time will build upon prior cross-sectional,

correlational studies by delineating what contributes to

adherence trajectories over time. For example, if one par-

ticular ACT type is associated with the poorest adherence,

clinicians may wish to prescribe other forms of ACT that

patients prefer and are thus more likely to adhere to. This

is clinically feasible because to date, there are no data to

suggest differences in health outcomes based on which

ACT is used for patients with CF (Main, Prasad, &

Schans, 2005).

New statistical methods, including group-based trajec-

tory models (GBTM), provide the ability to identify adher-

ence sub-groups, including those with the poorest

adherence (e.g., highest risk). These methods have been

successfully applied to other areas and populations,

including criminology (Jennings & Piquero, 2008),

dental caries (Broadbent, Thomson, & Poulton, 2008),

American Indian youth (Stiffman, Alexander-Eitzman,

Silmere, Osborne, & Brown, 2007), and most recently

dyspnea symptoms in marine transportation workers

(Arrandale, Koehoorn, Macnab, & Kennedy, 2009); how-

ever, it has not been used in pediatric psychology research,

which was the central goal of the current study.

The purpose of this study was 3-fold: (1) to describe

rates of adherence to ACT over time for a sample of chil-

dren, adolescents and adults with CF using the Daily

Phone Diary (DPD), an empirically supported assessment

of adherence (Quittner, Modi, Lemanek, Ievers-Landis, &

Rapoff, 2008), (2) to identify and characterize adherence

trajectories for individuals with CF, and (3) to identify

Adherence Trajectories in Cystic Fibrosis 1029

at University of M

iami - O

tto G. R

ichter Library on February 13, 2014

http://jpepsy.oxfordjournals.org/D

ownloaded from

predictors of adherence trajectories. It was hypothesized

that a minimum of two distinct adherence trajectory sub-

groups (e.g., high and low) would be identified and that

these trajectories would be predicted from patient-specific

factors such as age, sex, airway clearance treatment, base-

line pulmonary functioning, and family income.

MethodsParticipants

This was a secondary analysis of data collected for a ran-

domized, prospective clinical trial examining differences in

efficacy of three types of ACT (e.g., PD&P, flutter device, or

Vest) on pulmonary functioning over a 3-year period.

Participants for the clinical trial included individuals with

CF who were 7 years of age or older across 20 CF centers.

All patients had a confirmed diagnosis of CF based on

either a sweat chloride or two documented CFTR muta-

tions and forced expiratory volume in 1 s predicted

(FEV1% predicted) of at least 45%. Exclusion criterion

included the following: hospitalization for a pulmonary

exacerbation, episode of gross hemoptysis (>249 ml)

within 60 days prior to screening, pneumothorax in the

6 months prior to screening, or unwillingness to be rando-

mized to one of three ACTs. Although a sample size of 180

patients was expected, difficulties with recruitment re-

sulted in a final sample of 166 participants for the larger

clinical trial. Fifteen participants withdrew within the first

60 days after randomization, 11 of whom withdrew on the

day of randomization. The clinical trial ended early because

withdrawal rates were significantly higher in PD&P than in

Flutter or HFCWO groups and in older age groups (Sontag

et al., 2010).

Adherence data were available for at least one time

point for 153 participants; however, there were missing

data (16%) for participants across the five assessment

points (pre-randomization: n¼ 130; 4-month: n¼ 131;

8-month: n¼ 121; 12-month: n¼ 115; and 16-month:

n¼ 100). Forty-three percent of participants had five com-

plete data assessment points. Participants in the study

ranged from 7 to 44 years of age at the time of enrollment

(M¼ 14.3 years, SD¼ 7.9 years), and 55% were male. A

majority of participants were Caucasian (86%), while 7%

were Hispanic, 3% were African–American, and 2% were

other. Baseline FEV1% predicted was 86.7% (SD¼ 19.1),

while the mean FEV1% predicted was 85.1 (SD¼ 18.8) for

participants at Time 5, suggesting mild disease severity at

both time points. A majority of participants were either

children or adolescents (80%) and the type of ACT was

equally distributed across participants (PD&P¼ 33%,

Flutter ¼ 30%, and Vest ¼ 37%).

Procedure

Detailed procedures and results regarding the larger rando-

mized trial are reported elsewhere (Sontag et al., 2010);

however, it is important to note that study consent was

obtained from caregivers of children 7–17 years of age and

adults 18 years and older while assent was obtained from

adolescents. Participants were given instructions and train-

ing on their assigned ACT, which was prescribed twice

daily for 20 min/session. For purposes of this secondary

data analysis, adherence was the primary variable of

interest. Adherence was measured at baseline and every

4 months thereafter. Because adherence data were sparse

after the fifth assessment and the study was discontinued

for most participants by the sixth assessment due to high

drop-out rate in the adult age group (Sontag et al., 2010),

data analyses were performed on the first five assessments.

Institutional Review Board Approval was received from all

20 sites participating in the clinical trial.

Measures

Daily Phone Diary

The DPD, a well-established empirically supported assess-

ment measure of adherence (Quittner et al., 2008), uses a

cued recall procedure to track parents, adolescents, and

adults through their activities over the past 24 hours and

provides a detailed analysis of activity patterns, compan-

ions, and mood (Quittner & Opipari, 1994; Quittner,

Opipari, Regoli, Jacobsen, & Eigen, 1992). For all activities

lasting 5 min or longer, participants (e.g., caregivers of chil-

dren 7–13 years of age, adolescents 14–18 years of age, and

adults 18 years and older) reported the type of activity,

duration, and who was present. The interviewer assisted

participants in reconstructing their day as accurately as pos-

sible by providing prompts, such as the time of day or in-

formation about the previous activity (after you finished

dinner, what did you do next?). As the phone diary was

conducted, each activity was recorded by the interviewer on

a computer screen with clock hands which rotated through

a 24-hour clock, a set of activities, companions, and a rating

of mood ranging from 1 (extremely negative) to 5 (extreme-

ly positive). A set of two DPDs (one weekday and one week-

end day) was conducted by phone at each assessment

point. The DPD has yielded reliable stability coefficients

over a 3-week period (r¼ 0.61–0.71, p < .01) and high

levels of inter-rater reliability (>90%) in a CF population

(Quittner et al., 1992). Strong convergent validity was

1030 Modi et al.

at University of M

iami - O

tto G. R

ichter Library on February 13, 2014

http://jpepsy.oxfordjournals.org/D

ownloaded from

found between adherence calculated through the DPD and

objective measures of adherence (Modi et al., 2006; Modi &

Quittner, 2006b). For the current study, time spent in ACT

was extracted from the DPD to measure the duration (in

minutes) each patient spent in ACT each day.

Demographic Questionnaire

A basic demographic questionnaire, which provided infor-

mation about child age, sex, and family income, was ob-

tained via a case-report form.

Health Status

Respiratory functioning was measured by pulmonary func-

tion tests (PFTs). Specifically, FEV1% predicted using

equations for age, sex, and weight were performed by a

trained technician at all sites.

Statistical Analyses

Data analysis proceeded in two discrete stages: a two-part

descriptive phase in which we first sought to understand

and describe the characteristics of the data in general,

and then identified and described trajectories for specific

subgroups within the broader sample of individuals with

CF; and an inferential phase in which we sought to identify

clinically significant predictors (e.g., age, baseline pulmon-

ary functioning) of these adherence trajectory subgroups.

All data were analyzed using SAS v9.1� (SAS Institute,

Cary, NC, USA) and the PROC TRAJ Macro, a closed

source module developed specifically for use with SAS soft-

ware (http://www.andrew.cmu.edu/user/bjones).

Descriptive Phase (Aim 1)

Univariate statistics, including frequency counts as well as

measures of central tendency, variability, and association,

were used to describe sample characteristics, medical his-

tory, and patterns of adherence for this sample. Adherence

rates were calculated using the number of minutes each

participant engaged in airway clearance over the 2-day

DPD period divided by the number of minutes prescribed

for each individual (e.g., 40 min daily was prescribed for all

participants). For example, if a patient conducted airway

clearance for 30 min on day 1 and 20 min on day 2,

the adherence rate would be as follows: [(30þ 20)/

80]� 100¼ 62.5%. Adherence rates were computed for

each assessment point and for the entire study period.

Group-based Trajectories (Aim 2)

A semi-parametric, group-based approach to model estima-

tion was used to identify and characterize ACT adherence

rates over time. This technique, known as GBTM, draws

from a broader class of mixture models in which a finite

number of subgroups within a broader population are

allowed to have different patterns (or trajectories) of

change over time (Nagin, 2005). GBTM is particularly

useful when the number of subgroups and shapes of the

respective trajectories are unknown. To this extent, a

two-step approach for model estimation was used. First,

two-, three-, and four-group solutions were compared to

identify the optimal number of trajectories based on model

Bayesian Information Criterion (BIC) values, where the

value closest to zero value indicates the best fit. Second,

parameter estimates and Wald-statistics for intercept-only,

linear, quadratic, and cubic polynomials were compared to

determine the preferred model. In this case, the preferred

model included three separate trajectories, which were

then plotted for visual inspection. Assignment of indi-

viduals to the three trajectory subgroups was based

on the highest posterior probability estimate for the

final, best-fitting model. A censored normal probability

distribution served as the reference distribution for

all three curves analyzed here. For linear trajectories, a

mixed model was tested to examine the slope of adherence

over time.

Inferential Phase (Aim 3)

In addition, five cumulative logistic regression models

(i.e., ordinal logistic regression) were specified and tested

to identify patient-specific factors capable of distinguishing

the three adherence trajectory subgroups. Cumulative lo-

gistic regression was chosen due to the ordinal nature of

the adherence groups. Although the specific number of

adherence groups was initially unknown, it was reasonable

to assume that the groups would be ordinal in nature (e.g.,

low, medium, high adherence). The proportional odds

model of cumulative logistic regression has the added

advantage of offering summary odds ratios for each of

the predictors rather than multiple odds ratios for each

pair of outcomes (Gameroff, 2005). The outcome of inter-

est was trajectory group membership. Testable covariates

included patient sex, age, randomized ACT type, baseline

FEV1% predicted, and household income.

ResultsDescriptive Phase (Aims 1 and 2)

Descriptive data, including means and standard deviations

for adherence rates across the five time points, are as

follows: pre-randomization: M¼ 36.1 (SD¼ 30.4);

Adherence Trajectories in Cystic Fibrosis 1031

at University of M

iami - O

tto G. R

ichter Library on February 13, 2014

http://jpepsy.oxfordjournals.org/D

ownloaded from

4-month: M¼ 57.5 (SD¼ 37.2); 8-month: M¼ 54.5

(SD¼ 41.5); 12-month: M¼ 58.0 (SD¼ 42.8);

16-month: M¼ 55.9 (SD¼ 35.6). Average adherence

rates across all time points were slightly better than

50% (range 36–58%), and were consistently better than

50% after randomization. A spaghetti plot of adherence

data is presented in Figure 1A to demonstrate inter and

intra-patient variability.

GBTM analyses were conducted on two-, three-, and

four-group models, with time as the only covariate repre-

senting the five different observation periods. All five time

points were used, including baseline values, specifically to

obtain a more complete description of adherence patterns

over time for individuals with CF. All three models are

presented in Table I to illustrate why the three-group

model was preferred. Results indicated that a three-group

model best explained the data compared with a two- and

four-group model using BIC values as our criterion for

statistical significance. Table II demonstrates the best-

fitting solution for a three-group model included a linear

trajectory that assigned 14% of patients to a low-adherence

group (Group 1), a cubic trajectory for the medium-

adherence group (Group 2 accounting for 49% of the

total sample), and a quadratic trajectory for the

Visits

Adh

eren

ce R

ate

0

20

40

60

80

100

120

140

160

Visits

Adh

eren

ce R

ate

0

20

40

60

80

100

120

Visits

Adh

eren

ce R

ate

0

20

40

60

80

100

120

140

160

Visits

Adh

eren

ce R

ate

0

20

40

60

80

100

120

140

160

Visits

1 2 3 4 5

1 2 3 4 5

1 2 3 4 5

1 2 3 4 5

1 2 3 4 5

Adh

eren

ce R

ate

0

20

40

60

80

100

120

140

160

A

B

C

D

E

Figure 1. (A) Individual trajectories of adherence over time, (B) Trajectories of individuals in the high adherence group, (C) Trajectories

of individuals in the medium adherence group, (D) Trajectories of individuals in the low adherence group and (E) Final estimates of each

adherence trajectory subgroup.

Table I. BIC values and predicted group percents for GTBM for two-,

three-, and four-group trajectory solutions

Predicted group percents

Model

Number

of groups BIC 1 2 3 4

1 2 �2627.19 40.3 59.7

2 3 �2607.76 15.1 48.5 36.5

3 4 �2626.16 15.6 48.1 36.3 <1.0

Bold represents the final model that was selected.

1032 Modi et al.

at University of M

iami - O

tto G. R

ichter Library on February 13, 2014

http://jpepsy.oxfordjournals.org/D

ownloaded from

high-adherence group (Group 3 accounting for 37% of the

total sample). Group membership for each trajectory can

be found in Figures 1B through D. Individual trajectories

for the high-adherence Figure 1B, medium-adherence

Figure 1C, and low-adherence Figure 1D groups are

plotted. Final estimates of the trajectories are presented

in Figure 1E. Results indicated a significant negative

slope for the low-adherence group [F(4, 94)¼ 3.22;

p < .05].

Predictors of Adherence Trajectories (Aim 3)

Using the newly created adherence trajectories (low,

medium, and high), patient-specific factors that predicted

group membership were examined. Of the five cumulative

single covariate logistic regression models specified, only

ACT type emerged as a statistically significant predictor of

adherence group membership (Table III). Specifically, we

found significantly more participants using the Flutter

device in the medium-adherence group (see Table IV).

A trend was also noted for family income, in which

patients with a family income greater than $50,000 were

1.87 times more likely to be in the high-adherence trajec-

tory group. Similarly, a positive trend was noted for pul-

monary functioning, in which patients with better FEV1%

predicted were slightly more likely to be in the high adher-

ence group.

Discussion

Results from the current study build upon prior

cross-sectional research examining adherence to ACT for

patients with CF and suggest significant inter-patient vari-

ability. Prior to randomization for the ACT clinical trial,

adherence to this treatment was quite low at �36% for

the entire sample. This translates into patients performing

14 min of ACT each day compared to the recommended

40 min per day. However, as expected, adherence to ACTs

increased with enrollment in a clinical trial to 58%.

Although this rate is still low, it is consistent with prior

studies (Abbott et al., 1996; Modi et al., 2006; Myers &

Horn, 2006; Quittner et al., 2000) and suggests that pa-

tients with CF have poor adherence to this critical aspect of

their treatment regimen. Longitudinal results also indi-

cated that rates of adherence remained relatively stable

over the 16-month study period, indicating that patients

with CF who enroll and remain in a randomized, clinical

trial are able to consistently engage in ACT �50% of the

recommended time.

One novel aspect of this study was its examination of

trajectories of adherence over time in a large cohort of

patients with CF. Data indicated three distinct groups,

including a low, medium, and high adherence group.

Specifically, it appears that patients whose initial adher-

ence was lowest (e.g., 17%) remained low over time.

In contrast, patients who exhibited the highest rate of

adherence initially maintained higher levels of adherence

over time. The medium adherence group, which encom-

passed the largest proportion of individuals with CF

(49%), displayed the most variability in adherence. As sug-

gested by prior studies, this is likely due to ACT barriers,

such as competing activities and forgetting (Abbott et al.,

1996; Modi & Quittner, 2006a), which interfere at various

Table III. Logistic regression: patient predictors of adherence

trajectories

Variable OR CI0.95 p-value

Sex (male as reference category) 1.26 0.69, 2.31 0.46

Age 0.99 0.96, 1.03 0.72

ACT type 0.04

PD&P (reference group)

Flutter 2.46 1.13, 5.36 0.02

Vest 1.08 0.52, 2.23 0.84

Baseline FEV% predicted 1.05 0.99, 1.03 0.06

*Household income

(<$50,000 as

reference category)

1.87 0.92, 3.80 0.08

Note. All models tested met the score test for the proportional odds assumption.

*Multiple iterations were used to examine household income differences,

including a four- and a seven-category income variable with results indicating

no statistically significant relationship.

Table II. Final three-group model for adherence trajectories

Group Parameter Estimate (SE) T p

Low Intercept 17.73 (11.22) 1.58 0.11

Linear model �8.58 (4.13) �2.08 0.03

Medium Intercept �43.82 (25.06) �1.75 0.08

Cubic model 3.61 (1.28) 2.82 <0.01

High Intercept 0.73 (11.42) 0.06 0.94

Quadratic model �8.22 (1.49) �5.53 <0.01

Table IV. ACT type by adherence trajectory subgroup

Group PD&P Flutter Vest

Low adherence (n¼ 22) 7 (14%) 7 (15%) 8 (14%)

Medium adherence (n¼ 75) 20 (40%) 31 (67%) 24 (42%)

High adherence (n¼ 56) 23 (46%) 8 (17%) 25 (44%)

Adherence Trajectories in Cystic Fibrosis 1033

at University of M

iami - O

tto G. R

ichter Library on February 13, 2014

http://jpepsy.oxfordjournals.org/D

ownloaded from

times and occur inconsistently. It is possible that individ-

uals in the high adherence group are better able to problem

solve their barriers to ACT adherence and incorporate

these strategies into daily routines compared to individuals

in the medium adherence group. Overall, these adherence

trajectories allowed for identification of patients who are

highest risk for poor adherence by demonstrating that

patients who exhibit adherence rates below 25% are

likely to maintain these levels over time, placing them at

risk for increased morbidity (e.g., pulmonary exacerba-

tions, acquisition of drug-resistant microorganisms). It is

this subgroup that would benefit from the most intensive

intervention to improve adherence. However, even the high

and medium groups are below 50% and thus would bene-

fit from adherence interventions. Specifically, adherence

interventions that focus on providing education to fill

gaps in knowledge, teaching and ensuring that patients

have the skills to conduct the therapy, identifying and

problem solving around barriers to adherence, and match-

ing the type of therapy that best suits the needs of the

patient and their family are likely the most beneficial

(Homnick, 2007; Kahana et al., 2008).

Interestingly, ACT was the only significant patient-

specific predictor of adherence trajectories. Patients pre-

scribed Flutter were more likely to be in the medium

adherence group compared to PD&P and Vest. The

Flutter can be completed independently once the tech-

nique has been mastered, which is likely desirable for ado-

lescents and adults. However, because the Flutter requires

active participation (versus passive with Vest), patients may

become bored or disinterested in the therapy for the full

prescribed duration. These findings suggest that patients

prescribed the Flutter device are not likely to have high

rates of adherence over time. If, in contrast, patients pre-

scribed PD&P or Vest are more likely to be in the

high-adherence trajectory, it is important to incorporate

patient preference when making decisions about ACTs,

especially given lack of evidence regarding efficacy differ-

ences between the three ACT methods (Main et al., 2005).

Homnick (2007) has suggested the need to adapt ACTs

based on disease severity and patient preference to encour-

age optimal adherence, with frequent monitoring to ensure

health status does not decline.

While age and sex were not associated with adherence

trajectories, a trend was noted for both family income and

baseline pulmonary functioning. Specifically, patients with

CF were almost twice as likely to be in the high adherence

group if their family income was greater than $50,000.

These results are similar to studies in pediatric epilepsy

(Modi et al., 2008) and liver transplant (Berquist et al.,

2006) suggesting a positive association between adherence

and socioeconomic status. In addition, patients with CF

with higher pulmonary functioning trended toward being

in the high adherence group compared to the low adher-

ence group. Future studies with more variability in pul-

monary functioning (e.g., patients with more severe

disease) would better elucidate these findings.

While this study is the first to examine longitudinal

adherence to ACT for children, adolescents, and adults

with CF, several limitations exist that provide directions

for future research. The identification of adherence trajec-

tories and predictors of these trajectories represented the

first step in understanding adherence behaviors over time.

An important next step is to examine the relations between

adherence trajectories and health outcomes, including

pulmonary functioning, nutritional status, and health-

related quality of life. With this information, clinical

researchers may be able to identify optimal levels of adher-

ence, which promote the best outcomes. Future studies

should also include a larger cohort of adults with CF

to elucidate developmental differences in adherence.

Although the use of GBTM techniques was a major

strength of this study, research from other fields has sug-

gested that large sample sizes are necessary to identify

trajectories. While our sample size fell below the suggested

200 cases, studies have successfully used sample sizes con-

sistent with ours (Conklin et al., 2005; Mulvaney,

Lambert, Garber, & Walker, 2006). An important area

for future research is a confirmatory analysis of the low,

medium, and high adherence trajectories we identified.

Finally, these data are a secondary analysis of a randomized

clinical trial examining differences between three ACT

types and thus must be considered within this context.

Sontag et al. (2010) have reported lessons learned from

this trial, including early termination due to differential

drop-out in the adult PD&P study arm, which contributed

to inadequate power to detect group differences. While the

early termination of the trial has important implications for

adherence, because adherence was the primary outcome

variable, we had sufficient longitudinal data to compute

our results.

Overall, this is the first study to use GBTM to examine

adherence behaviors in individuals with a chronic illness

that requires a time-consuming and complex treatment

regimen. Adherence trajectories, such as the ones identi-

fied in this study, are likely to generalize to other chronic

illness populations with complex treatment regimens,

including diabetes, sickle cell disease, and asthma and

1034 Modi et al.

at University of M

iami - O

tto G. R

ichter Library on February 13, 2014

http://jpepsy.oxfordjournals.org/D

ownloaded from

should be evaluated in these populations. Examining ad-

herence trajectories holds promise for identifying the level

of adherence necessary to obtain the best health outcomes

and quality of life.

Acknowledgements

The authors thank all of the children, adolescents, adults,

and their families who participated in this study. In addi-

tion, we would like to thank all the research assistants who

conducted the DPDs. We would also like to acknowledge

the following institutions, Investigators and Coordinators

of the Airway Secretion Clearance Trial: Baylor College of

Medicine (Texas Children’s Hospital), Oermann, Chris,

Morris, Lisa; Tulane University School of Medicine,

Davis, Scott, Broussard, Annette; Baystate Medical

Center, Gerstle, Robert, Pellett, Amira; Schneider

Children’s Hospital/Long Island Medical Center, DeCelie-

Germana, Joan, Hurley, Jody; University of Florida,

Wagner, Mary, Capen, Cindy; Children’s Healthcare of

Atlanta at Scottish Rite, Scott, Peter, Crews, Barbara;

New England Medical Center, Yee, William, Ulles,

Monica; University of California, San Diego, Pian, Mark,

Boone, Debbie; Children’s Hospital of Buffalo, McMahon,

Colin, Prentice, Sandy; Programa Pediatrico Pulmonar de

San Juan, Rodriguez-Santana, Jose, Medina, Vivian;

Connecticut Children’s Medical Center, Lapin, Craig,

Drapeau, Ginny; The University of Texas Southwestern

Medical Center at Dallas, Gelfand, Andy, Urbanczyk,

Brenda; St. Vincent’s Hospital and Medical Center of

New York, Berdella, Maria, Grece, Caroll Anne;

University of Rochester, Voter, Karen, Sierzega, Renata;

Indiana University, Eigen, Howard, Blagburn, Mary;

University of Wisconsin, Rock, Michael, Yngsdal-Krenz,

Rhonda; Wake Forest University, Schechter, Michael,

Hunt, Meagan; Kaiser Permanente Medical Care Program,

Shay, Greg, Farmer, Gail; University of Nebraska Medical

Center, Colombo, John, Acquizzino, Dee; St. Louis

University, Albers, Gary, Lewis, Pat.

Funding

Hill-Rom Inc. (formerly American Biosystems, Inc.) and

the United States Cystic Fibrosis Foundation.

Conflict of interest: American Biosystems provided one of

the three ACT methods used in the larger randomized

clinical trial. Dr. Accurso has served on the advisory

board for American Biosystems.

References

Abbott, J., Dodd, M., & Webb, A.K. (1996). Health

perceptions and treatment adherence in adults with

cystic fibrosis. Thorax, 51, 1233–1238.

Arias Llorente, R.P., Bousono Garcia, C., & Diaz

Martin, J.J. (2008). Treatment compliance in children

and adults with cystic fibrosis. Journal of Cystic

Fibrosis, 7, 359–367.

Arrandale, V.H., Koehoorn, M., Macnab, Y.,

& Kennedy, S.M. (2009). Longitudinal analysis of

respiratory symptoms in population studies with a

focus on dyspnea in marine transportation workers.

International Archives of Occupational and

Environmental Health, 82, 1097–1105.

Berquist, R.K., Berquist, W.E., Esquivel, C.O., Cox, K.L.,

Wayman, K.I., & Litt, I.F. (2006). Adolescent

non-adherence: prevalence and consequences in

liver transplant recipients. Pediatric Transplant, 10,

304–310.

Bradley, J.M., Moran, F.M., & Elborn, J.S. (2006).

Evidence for physical therapies (airway clearance and

physical training) in cystic fibrosis: an overview of

five Cochrane systematic reviews. Respiratory

Medicine, 100, 191–201.

Broadbent, J.M., Thomson, W.M., & Poulton, R. (2008).

Trajectory patterns of dental caries experience in the

permanent dentition to the fourth decade of life.

Journal of Dental Research, 87, 69–72.

Conklin, C.A., Perkins, K.A., Sheidow, A.J., Jones, B.L.,

Levine, M.D., & Marcus, M.D. (2005). The return

to smoking: 1-year relapse trajectories among

female smokers. Nicotine and Tobacco Research, 7,

533–540.

Conway, S.P., Pond, M.N., Hamnett, T., & Watson, A.

(1996). Compliance with treatment in adult patients

with cystic fibrosis. Thorax, 51, 29–33.

Cystic Fibrosis Foundation (2009). About Cystic Fibrosis,

Retrieved from http://www.cff.org/AboutCF/

Gameroff, M.J. (2005). Using the proportional odds

model for health-related outcomes: Why, when, and

how with various SAS� procedures. In: Proceedings

of the 30th annual SAS User’s Group International

Conference. Retrieved from http://www2.sas.com/

proceedings/sugi30/205-30.pdf

Homnick, D.N. (2007). Making airway clearance success-

ful. Paediatric Respiratory Reviews, 8, 40–45.

Jennings, W.G., & Piquero, A.R. (2008). Trajectories of

non-intimate partner and intimate partner homicides,

Adherence Trajectories in Cystic Fibrosis 1035

at University of M

iami - O

tto G. R

ichter Library on February 13, 2014

http://jpepsy.oxfordjournals.org/D

ownloaded from

1980-1999: The importance of rurality. Journal of

Criminal Justice, 36, 435–443.

Kahana, S., Drotar, D., & Frazier, T. (2008). Meta-

analysis of psychological interventions to promote

adherence to treatment in pediatric chronic health

conditions. Journal of Pediatric Psychology, 33,

590–611.

Latchford, G., Duff, A., Quinn, J., Conway, S.,

& Conner, M. (2009). Adherence to nebulised

antibiotics in cystic fibrosis. Patient Education and

Counseling, 75, 141–144.

Main, E., Prasad, A., & Schans, C. (2005). Conventional

chest physiotherapy compared to other airway

clearance techniques for cystic fibrosis. Cochrane

Database System Review, CD002011.

McNamara, P.S., McCormack, P., McDonald, A.J.,

Heaf, L., & Southern, K.W. (2009). Open adherence

monitoring using routine data download from an

adaptive aerosol delivery nebuliser in children with

cystic fibrosis. Journal of Cystic Fibrosis, 8, 258–63.

Modi, A.C., Lim, C.S., Yu, N., Geller, D., Wagner, M.H.,

& Quittner, A.L. (2006). A multi-method assessment

of treatment adherence for children with cystic

fibrosis. Journal of Cystic Fibrosis, 5, 177–185.

Modi, A.C., Morita, D.A., & Glauser, T.A. (2008).

One-month adherence in children with new-onset

epilepsy: white-coat compliance does not occur.

Pediatrics, 121, e961–966.

Modi, A.C., & Quittner, A.L. (2006a). Barriers to

Treatment Adherence for Children with Cystic

Fibrosis and Asthma: What Gets in the Way?

Journal of Pediatric Psychology, 31, 846–858.

Modi, A.C., & Quittner, A.L. (2006b). Utilizing com-

puterized phone diary procedures to assess health

behaviors in family and social contexts. Children’s

Health Care, 35, 29–45.

Mulvaney, S., Lambert, E.W., Garber, J., & Walker, L.S.

(2006). Trajectories of symptoms and impairment for

pediatric patients with functional abdominal pain: a

5-year longitudinal study. Journal of the American

Academy of Child and Adolesccent Psychiatry, 45,

737–744.

Myers, L.B. (2009). An exploratory study investigating

factors associated with adherence to chest phy-

siotherapy and exercise in adults with cystic fibrosis.

Journal of Cystic Fibrosis, 8, 425–427.

Myers, L.B., & Horn, S.A. (2006). Adherence to chest

physiotherapy in adults with cystic fibrosis. Journal

of Health Psychology, 11, 915–926.

Nagin, D. (2005). Group-based modeling of development.

Cambridge: Harvard University Press.

Passero, M.A., Remor, B., & Salomon, J. (1981). Patient-

reported compliance with cystic fibrosis therapy.

Clinical Pediatrics (Phila), 20, 264–268.

Patterson, J.M., Wall, M., Berge, J., & Milla, C. (2008).

Gender differences in treatment adherence

among youth with cystic fibrosis: development of a

new questionnaire. Journal of Cystic Fibrosis, 7,

154–164.

Quittner, A.L., & Opipari, L.C. (1994). Differential

treatment of siblings: interview and diary analyses

comparing two family contexts. Child Development,

65, 800–814.

Quittner, A.L., Opipari, L.C., Regoli, M.J., Jacobsen, J.,

& Eigen, H. (1992). The impact of caregiving and

role strain on family life: Comparisons between

mothers of children with cystic fibrosis and

matched controls. Rehabilitation Psychology, 37,

289–304.

Quittner, A.L., Drotar, D., Ievers-Landis, C.E.,

Seidner, D., Slocum, N., & Jacobsen, J. (2000).

Adherence to medical treatments in adolescents with

cystic fibrosis: The development and evaluation of

family-based interventions. In D. Drotar (Ed.),

Promoting Adherence to Medical Treatment in

Childhood Chronic Illness: Interventions and Methods

(pp. 383–407). Hillsdale, NJ: Erlbaum Associates, Inc.

Quittner, A.L., Modi, A.C., Lemanek, K.L.,

Ievers-Landis, C.E., & Rapoff, M.A. (2008).

Evidence-based assessment of adherence to medical

treatments in pediatric psychology. Journal of

Pediatric Psychology, 33, 916–936.

SAS Institute (Version 9.1). Cary, NC, USA.

Sontag, M.K., Quittner, A.L., Modi, A.C., Giles, D.,

Koenig, J.M., Oermann, C., et al. (2010). Lessons

learned from a randomized trial of airway secretion

clearance techniques in cystic fibrosis. Pediatric

Pulmonology, 45(3), 291–300.

Stiffman, A.R., Alexander-Eitzman, B., Silmere, H.,

Osborne, V., & Brown, E. (2007). From early to late

adolescence: American Indian youths’ behavioral

trajectories and their major influences. Journal of the

American Academy of Child and Adolescent Psychiatry,

46, 849–858.

White, D., Stiller, K., & Haensel, N. (2007). Adherence

of adult cystic fibrosis patients with airway clearance

and exercise regimens. Journal of Cystic Fibrosis, 6,

163–170.

1036 Modi et al.

at University of M

iami - O

tto G. R

ichter Library on February 13, 2014

http://jpepsy.oxfordjournals.org/D

ownloaded from

White, T., Miller, J., Smith, G.L., & McMahon, W.M.

(2009). Adherence and psychopathology in children

and adolescents with cystic fibrosis. European Child

and Adolescent Psychiatry, 18, 96–104.

Zindani, G.N., Streetman, D.D., Streetman, D.S.,

& Nasr, S.Z. (2006). Adherence to treatment in

children and adolescent patients with cystic fibrosis.

Journal of Adolescent Health, 38, 13–17.

Adherence Trajectories in Cystic Fibrosis 1037

at University of M

iami - O

tto G. R

ichter Library on February 13, 2014

http://jpepsy.oxfordjournals.org/D

ownloaded from