Journal of Pediatric Psychology, Vol. 27, No. 1, 2002, pp. 67-76
© 2002 Society of Pediatric Psychology
Daily Reports and Pooled Time Series Analysis: Pediatric Psychology Applications
1 Washington State University Vancouver, 2 Oregon Social Learning Center, 3 Oregon Health Sciences University
All correspondence should be sent to Elizabeth Soliday, Psychology Program, WSU Vancouver, 14204 NE Salmon Creek Ave., Vancouver, Washington 98686. E-mail: soliday{at}vancouver.wsu.edu .
 |
Abstract |
|---|
 Top Abstract IntroductionMethodResultsDiscussionConclusionsReferences |
|---|
Objective: To apply daily reports and pooled time series analysis (PTSA) to issues in pediatric psychology research. We discuss specific applications for this procedure in analyzing repeated observations for a small sample, including medication effects, caregiving role strain, pain reports, and treatment effects.
Methods: In the PTSA example presented, 20 daily behavior reports were provided by parents of 10 children with steroid-sensitive nephrotic syndrome (SSNS) during high-dose steroid administration and tapering.
Results: The full model, including child age, medication dosage, and between-subjects effects, significantly predicted children's aggressive behavior and anxious/depressed behavior. Steroid dosage significantly predicted aggressive, but not anxious/depressed, behavior.
Conclusions: Daily reports analyzed using PTSA provided insight into serious behavioral side effects of steroid medications used to treat SSNS. We discuss the role of pediatric psychologists in addressing medication side effects and other time-related effects detectable using this methodology.
Key words: pooled time series analysis; within-subjects design; CBCL; prednisone; nephrology.
|
Introduction |
|---|
|
TopAbstractIntroductionMethodResultsDiscussionConclusionsReferences |
|---|
Many pediatric psychologists work with small numbers of children who have rare, chronic medical disorders. For example, pediatric illnesses such as cancer, cystic fibrosis, and certain gastrointestinal disorders have prevalence rates generally below 1% (Engstroem & Lindquist, 1998
; Miller, Jelalian, &
Stark, 1999; Powers, Vannatta,
Noll, Cool, & Stehbens, 1995). Similarly, pediatric kidney
disease, an example of which we use here to demonstrate methodological
techniques, affects fewer than 50 children per million annually. However, it
is generally agreed that, although rarely occurring, pediatric chronic
illnesses merit psychologists' attention due to their potential impact on
children's developmental experiences and overall functioning. Whereas over two
decades of comparing chronically ill and healthy children's adjustment have
informed the research and clinical community on the general impact of
pediatric chronic illness, researchers have begun to acknowledge the
usefulness of within-subjects approaches in delineating the complexities of an
individual's psychological health (e.g.,
Bolger, Delongis, Kessler, & Schilling,
1989; West & Hepworth,
1991). In this article, we discuss the unique and valuable
contribution that within-subjects approaches using repeated temporal
assessments can make in several timely pediatric psychology issues,
particularly those related to chronic illness.
In chronic disorders such as kidney disease, cancer, cystic fibrosis,
diabetes, migraine headaches, asthma, and inflammatory bowel disease, physical
symptoms tend to wax and wane over the course of the illness (e.g.,
Engstroem & Lindquist,
1998; Greene, Blanchard, &
Wan, 1994; Miller et al.,
1999). The reasons for this variability often remain unknown or
are poorly understood. Similarly, treatments for these disorders, rather than
being continuous, may be administered in "bursts" corresponding to
symptom change or disease relapse. Thus, assessments of children's functioning
conducted at one or even two time points, though useful for understanding
global adjustment, tell little about the impact of fluctuating illness and
treatment influences on children's psychosocial adjustment. An alternative
approach, repeated measurement of daily functioning, can provide insight into
the unique process of psychosocial adjustment over time when children have
chronic illness or psychological disturbance (e.g.,
Moore, Osgood, Larzelere, &
Chamberlain, 1994; Quittner,
Opipari, Regoli, Jacobsen, & Eigen, 1992).
In addition to providing greater insight into the complex and unique
adjustment challenges associated with chronic physical or psychological
illness, repeated temporal assessments and consequent appropriate analysis
offer considerable methodological advantages. Cook and Campbell
(1979) suggest that these
techniques help researchers control for history and maturation, both serious
threats to the internal validity of a longitudinal study. Learning more about
children's and families' adjustment over time may also prove extremely useful
in developing treatment recommendations. By gathering multiple temporal
assessments at theoretically and clinically meaningful time points,
researchers can better identify specific periods when children are at greatest
risk for difficulties and, relatedly, when they may most need intervention.
Depending on the specific issue, a repeated measures approach may be more
useful and valid for treatment planning than data from global assessments,
which may be subject to greater bias than daily reports (e.g.,
Patterson, 1982).
Unfortunately, until recently the complexity of data sets collected in
applied settings containing daily reports was often difficult to analyze. For
example, ordinary time series (which has occasionally been recommended)
generally assumes more data points than are usually available to clinical
researchers (e.g., a minimum of 50 data points). In addition, ordinary time
series cannot generalize because it is limited to single cases. Although
typical analyses of variance (ANOVA) models can partition variance into
between-subjects and within-subjects components, most often they cannot be
applied to complex time-ordered research designs such as the one presented
here, in which there is variable timing of the independent variable and a high
number of observations for each subject. Finally, single-subject analytic
techniques (i.e., visual analysis) can become overwhelming when attempting to
evaluate continuously collected data highly variable over time. Even though
some have argued that nothing clinically useful can be derived from
single-subject data not interpretable by visual analysis (e.g.,
Baer, 1977), Moore and his
colleagues were able to show clinically significant patterns in such data that
are masked by the variability (Moore et
al., 1994).
The purposes of this article are to (1) introduce an underutilized analytic procedure for repeated measures with small sample sizes, that is, pooled time series analysis (PTSA); (2) present examples of instances in which repeated temporal measures of behavior and/or adjustment may be a useful methodology for pediatric psychology researchers; and (3) present an example from our research on the behavioral side effects of steroid medications to demonstrate the utility of collecting daily reports and analyzing them using PTSA.
Analyzing Daily Reports With Pooled Time Series Analysis
PTSA relies on a regression approach to examine time-related trends,
offering the researcher the possibility of combining data from multiple
measurement points from several subjects to examine general time-related
effects (Jaccard & Wan,
1993, Ostrom,
1990). Daily reports analyzed via PTSA may offer several
considerable advantages to pediatric psychology researchers. In comparing PTSA
to MANOVA repeated measures models (which are mathematically related to
regression), group means are compared at different time points, and individual
differences may produce considerable variability in group means. This
variability can lead to inflated error terms and reduced power of statistical
tests, of particular concern for small samples
(Jaccard & Wan, 1993).
Moreover, because PTSA can partial out between-subject variance (i.e.,
individual differences) and use only within-subject variation, the timing of
interventions (i.e., changes in the independent variable) can vary across
subjects. Therefore, PTSA can control for history by exploiting a key feature
of the replicated time-series design
(Moore et al., 1994).
Furthermore, in contrast to the exclusive reliance on statistical
significance when researchers use MANOVA and similar traditional statistical
approaches, the regression coefficients provided by PTSA give researchers more
clinically useful effect sizes (Moore et
al., 1994). PTSA also offers the possibility of retaining data in
analyses for participants who may not have participated at each measurement
point (Allison, 1994), which is
common in clinical studies. Finally, PTSA allows researchers to correct for
serial dependence (described in greater detail later), whereas these
corrections are not available in traditional MANOVA repeated measures
models.
Examples of Pediatric Psychology Applications of Daily Reports and
PTSA
Numerous pediatric psychology issues readily lend themselves to the PTSA
approach. For example, role strain in parents of pediatric patients with
low-incidence disorders (e.g., Quittner et
al., 1992) could be examined using PTSA. PTSA could be applied to
disease- and treatment-related pain studies such as those conducted by
Donaldson and Moinpour (1992)
and Richardson and McGrath
(1983). Time-ordered data can
help pediatric researchers explicate the bio-behavioral effects of contextual
variables such as psychosocial stress on chronic pediatric conditions,
including migraine headaches, asthma, and inflammatory bowel disease (IBD;
viz., Greene et al., 1994).
Compared to single-subject analyses, PTSA offers advantages to researchers
evaluating the effects of highly specialized treatment programs applied to
small numbers of children (e.g., Werle,
Murphy, & Budd, 1993): the ability to generalize treatment
effects is improved, interventions do not have to be timed such that level and
slope have an opportunity to show dramatic changes, and the researcher does
not have to wait for stable baselines or for the occurrence of the clinical
event of interest.
Finally, medication side effects can affect quality of life, which could
have important implications for clinical treatment decision making
(Beans, 1999;
Hathaway, Winsett, Milstead, Wicks, &
Gaber, 1996). In our research example, we use PTSA to examine
corticosteroids' impact on one dimension of pediatric patients' quality of
life: behavior. Of the few published studies on this issue, reports indicate
increased symptoms such as depression, anxiety, irritability, restlessness,
and sleep and memory disturbances (Bender,
Lerner, & Poland, 1991;
Harris, Carel, Rosenberg, Joshi, &
Leventhal, 1986; Satel,
1990). As steroid medications become more widely used in pediatric
conditions, psychologists have an important role in assessing the clinical
significance of their behavioral side effects.
Research Example
Our research example involves children with steroid-sensitive nephrotic
syndrome (SSNS). SSNS is a chronically relapsing disorder characterized by
massive urinary loss of protein (proteinuria) and total body edema. Most
children with SSNS have a chronically relapsing disease course. Relapses are
most frequently treated with high-dose steroid medications, tapered according
to improvements in the child's condition. About half of children with SSNS
relapse frequently, requiring several 8- to 12- week courses of steroids
throughout the year. While 95% of children with SSNS outgrow their disease
without long-term detrimental effects, the average duration is approximately
10 years (Warshaw, 1994).
|
Method |
|---|
|
TopAbstractIntroductionMethodResultsDiscussionConclusionsReferences |
|---|
Participants
Parents of children with the diagnosis of SSNS were recruited from the Pediatric Nephrology Clinic at Doernbecher Children's Hospital at Oregon Health Sciences University (OHSU). Fifteen English-speaking families with children ranging from 3 to 16 years old agreed to participate. Five additional families declined participation. Reasons for not participating included lack of time and discomfort completing psychological surveys. There were no apparent differences in age and sex of child, duration of diagnosis, or parent marital status between participants and nonparticipants. The majority of respondents were mothers (87%). Of the 15 parents agreeing to participate, 10 completed the study; 5 did not complete because their child was rediagnosed as steroid resistant (n = 1), or the child did not relapse during the 16-month course of the investigation (n = 4). Eight (80%) of the remaining 10 children were male; their mean age was 8.2 years (range: 2.9 to 16.5 years). Nine children were white, and one was African American. Hollingshead (1975
) ratings
available for nine of the ten children indicated one child in class I
(highest), and two children each in classes II, III, IV, and V. The study
protocol was approved by the institutional review boards at OHSU and
Washington State University. Informed consent was obtained from each child's
parent or guardian.
Measure and Procedure
To assess behavioral side effects of prednisone, the Child Behavior
Checklist (CBCL; Achenbach,
1991,
1992) was used. The 118-item
CBCL provides an age- and sex-standardized assessment of a child's
internalizing and externalizing behavior problems. We chose the CBCL for its
applicability to the wide age range of our sample, for its normative data, and
because it has been widely used in behavioral research. To minimize parental
fatigue, telephone assessments during relapse included only the
anxiety/depression and aggression subscales of the CBCL.
To obtain a baseline measure of the child's behavior, parents completed the full CBCL at a time when their child was in remission, off prednisone, or on low dose alternate day therapy (not more than 0.5 mg/kg/48h). All participants completed baseline assessments at least 6 weeks prior to relapse. At the initiation of daily prednisone for relapse (2mg/kg, divided two times a day), the research staff conducted a series of telephone calls to assess the child's behavior. A round of five consecutive daily telephone calls was initiated 2 days after starting full-dose prednisone and then repeated every 2 weeks for a total of four rounds of calls occurring during weeks 1, 3, 5, and 7 of therapy for relapse (20 calls total). The prednisone dose was decreased to 2 mg/kg every other day (single a. m. dose) at the time of urinary remission and then tapered approximately 0.5 mg/kg approximately every 2 weeks thereafter. Thus, the timing of behavioral assessments corresponded to the child's tapering medication schedule, depending on the timing of each patient's urinary remission. This prospective, repeated measures study design allowed each child to act as his or her own control (baseline vs. relapse behavior) and allowed assessment of dose-related changes in each child's behavior.
|
Results |
|---|
|
TopAbstractIntroductionMethodResultsDiscussionConclusionsReferences |
|---|
Analyses
We used two methods for analyzing our study data. For purposes of comparing across subjects of varying ages and gender, T scores were used in all analyses. The first procedure involved analyzing individual participant effects by plotting each individual's behavior scores by dosage/time. Figures 1,1 and 2,2 present examples of graphed scores from four individual subjects. In Figure 1,Figure 1, the linear effect of decreasing behavior problems corresponding to decreasing steroid dosage is apparent for both subjects. Figure 2,Figure 2 presents two subjects for whom, on visual inspection, there appears to be no clearly identifiable pattern of effects between dosage and behavior problems.
|
|
|
|
In addition, mean scores for high-dose and steroid-free periods were calculated for each individual participant by collapsing the five assessments conducted in each period. During the relapse episode when children were on steroid medication, 70% (n = 7) had one or more days when their behavioral symptoms were borderline or exceeded clinically symptoms levels (scores were at or above the 95th percentile). Mean scores decreased from high-dose to steroid free periods in five cases (50%), as predicted; scores increased slightly for one case, which is opposite of the predicted direction; scores remained stable from high-dose to steroid-free periods in the remaining four (40%) cases. (The preceding information on mean scores is provided for descriptive purposes only. The reader should be aware that collapsing across mean scores is somewhat antithetical to the visual analysis and PTSA approaches.)
The second analytic procedure involved examining behavior during relapse
for the entire group of subjects over time, accomplished by conducting PTSA.
As discussed previously, PTSA relies on a regression model in which serial
observations, in this case, daily reports of child behavior, are combined
(pooled) from the entire sample. This procedure allowed us to examine the
influences of steroid dosing over
time.1 Additionally,
we were able to control for the effect of age (by entering it first in the
model), which was necessary due to the wide age range of our sample. Before
presenting results of these specific analyses, we review the issues of primary
concern, followed by our results. The following sections are aimed at
providing a general overview of the procedure; more detailed technical
discussion of PTSA is available in Allison
(1994), Ostrom
(1990), and Sayrs
(1989).
Data Structure. In PTSA, data from individual time points are
treated as individual cases. For example, the 10 individual children in our
study had behavior scores from 20 different time points. In PTSA, each time
point for each participant is treated as an individual case; thus, in our
example, 10 individuals x 20 measurement points equal 200
"cases." Thus, the data would be structured such that the
researcher would have 200 rows, each with a single variable, rather than 10
rows with 20 variables each (the typical format for comparison of means). In
contrast to other techniques, the data structure of PTSA has two particular
strengths. First, because it uses a sample of subjects, fewer observations per
participant are required than for more familiar strategies, such as
traditional time-series. Second, greater statistical power can be developed,
even when using small clinical samples, by capitalizing on several
observations for each participant. (Because much clinical data have positive
autocorrelation, effective sample size is recommended at somewhere between
N and N x O.) One may use data transformation
procedures such as TRANSPOSE in SPSS (SPSS
Inc., 1994) for existing data sets. Each individual case,
identified with a subject code, is then entered into a regression equation.
Any data analytic program with linear regression capabilities (e.g., SPSS,
SAS), can be used to analyze time-related trends
(Ostrom, 1990).
Between-Subjects Effects. In PTSA, individual differences are addressed using dummy codes. The total number of dummy codes necessary is n - 1; for example, in our study example with n = 10 participants, nine individual difference dummy codes are necessary. These dummy codes absorb overall differences across participants and thereby limit any substantive analysis to change over time, independent of individual differences. Thus, results from PTSA ensure that preexisting individual differences cannot account for any substantive results. (For description on use of dummy coding, see Stephens, 1992.)
Serial Dependence. One of the concerns arising in the use of PTSA
(or any other procedure including three or more time-sequential measures;
Allison, 1994) is observations
closer together in time will be more highly correlated than those farther
apart (Ostrom, 1990). However,
traditional regression analysis requires that error terms be independent, and
a fundamental problem in analysis of temporal data is violation of the
assumption of independent residuals
(Jaccard & Wan, 1993). The
effect of this phenomenon, known as serial dependence, is that it can
artificially inflate significance estimates. Several available procedures can
help correct serial dependence in PTSA.
Perhaps the simplest procedure is to run the regression analysis without
correction for serial dependence to obtain an autocorrelation estimate. SPSS,
for example, provides an estimate of serial dependence (autocorrelation). This
estimate, the Durbin-Watson statistic, is available on the /STATISTICS
subcommand of the REGRESSION command. The Durbin-Watson statistic can be
transformed to an autocorrelation estimate, r, calculated as:
 |
 |
;
Morrison, 1983;
Neter, Wasserman, & Kutner,
1985.) Subsequent regressions are run as "no
intercept" (/ORIGIN subcommand under REGRESSION command in SPSS)
regressions. A constant is calculated (1 - r) and is entered as a
predictor. According to Ostrom
(1990), two iterations of this
procedure should be sufficient to correct for serial dependence. It is also
possible to test the significance of the Durbin-Watson value
(Neter et al., 1985) to decide
whether to proceed with further iterations. Although the variable
transformation procedure can be tedious, this correction procedure for serial
dependence ultimately leads to more trust-worthy results
(Neter et al., 1985;
Ostrom, 1990). PTSA Results on Aggressive Behavior. To control for serial dependence, a first multiple regression was run with the following predictors: child's age (as a control variable), medication dosage (milligrams prednisone/child's weight in kilograms), and dummy codes for individual differences. The Durbin-Watson statistic was used to compute serial dependence (autocorrelation). After transforming variables as described above, including a y-intercept, subsequent iterations of the procedure were conducted until an autocorrelation estimate of -.15 resulted.
Our final test of this model (child's age, medication dosage, and individual differences as predictors of children's aggressive behavior) was statistically significant, F(11, 179) = 1039.43, p = 000. In this final equation, the regression weight (B) for medication dosage was 2.48, t (179) = 2.89, p =.004. The regression weight indicated that each single unit increment in prednisone dosage (one milligram per kilogram) corresponded to an increase of approximately 2.5 units of aggressive behavior. Parents rated 19 aggressive behavior items as occurring "never," "sometimes," or "frequently" on each data collection day. Thus, increases in aggressive behavior units could correspond to increases in either the overall number of behaviors occurring "sometimes" or "frequently" or an increase in the frequency of occurrence of behaviors previously occurring.
In addition, R2 values from the original regression were examined to assess the effects of the control variable (child age) and individual difference codes (R2 values from the no-intercept regression could not be interpreted in this case due to variance explained by the transformed constant, which was entered on the first step). Child age did not significantly predict aggressive behavior scores, Fchange (1, 198) = 1.64, p =.202. Individual differences did significantly predict aggressive behavior scores, Fchange (10, 189) = 10.84, p =.000.
PTSA Results on Anxious/Depressed Behavior. The same procedures used to correct serial dependence in aggressive behavior scores were applied to anxious/depressed behavior. After subsequent iterations, an autocorrelation estimate of.27 was achieved. The final test of this model (child's age, medication dosage, and individual differences as predictors of children's anxious/depressed behavior) was statistically significant, F(11, 179) = 1430.02, p = 000. In this final equation, the regression weight (B) for medication dosage was.69, t (179) = 1.56, p =.12. Although not statistically significant, the regression weight indicated that each single unit increment in prednisone dosage corresponded to an increase of approximately.69 units of anxious/depressed behavior. Child age did not significantly predict anxious/depressed behavior scores, Fchange (1, 198) = 1.64, p =.202. Individual differences significantly predicted anxious/depressed behavior scores, Fchange (10, 189) = 8.47, p =.000.
|
Discussion |
|---|
|
TopAbstractIntroductionMethodResultsDiscussionConclusionsReferences |
|---|
The within-subjects methodology of using daily behavior reports and PTSA in examining pediatric psychology issues can lead to conclusions relevant to pediatric psychologists and other clinicians. For example, the study example here demonstrated that the CBCL, which has a narrow range of measurement error (Achenbach, 1991
), proved to be
a useful tool in detecting changes in the 20 daily assessments with each
participant. The study design, in which data from a small number of
participants could be integrated, is one of the strongest because it provides
a true assessment of change while controlling for individual differences,
history, and maturation (Cook &
Campbell, 1979). The statistical controls we employed help us have
more confidence in the results of our study example. Moreover, because PTSA
provides regression coefficients, we also have effect sizes directly related
to our clinical question of interest.
Clinicians working with pediatric patients who take prednisone often have
the impression of serious medication-related side effects. In our study, those
impressions were partially supported. PTSA results indicate that children
experience significant elevations in reported aggressive behavior related to
increased prednisone dosage. However, anxious/depressed behaviors were not
significantly predicted by medication dosage. Although previous studies have
reported prednisone-related psychological disturbances, they had
methodological problems that preclude drawing firm conclusions regarding
prednisone's behavioral effects. In particular, inadequate information on
dosage and duration of medication exposure make it difficult to understand the
precise nature of prednisone's potential dosage-related side effects
(Satel, 1990). Using
within-subjects repeated measures methodology, we were able to tease out the
nature of the increase in behavior problems relative to increase in steroid
dosage for two separate behavioral dimensions. Drawing these conclusions would
not have been possible had we elected to use assessments at only one or two
time periods.
At this point, our use of two different analytic techniques merits discussion. Whereas our PTSA results suggest a medication-related effect on behavior over time for the entire group, visual analysis of individual subject data graphed over each assessment period (examples presented in Figures 1,1 and 2,2) indicates that high-dose steroid administration had identifiable effects on the behavior of some participants, but not all. In fact, approximately 40% of participants had no clear pattern of dosage-related behavior change identifiable by graphic presentation. To some degree, our visual inspection supports the suspicion that group designs can mask important individual differences. However, one could erroneously conclude, based on graphs from individual subjects such as those in Figure 2,Figure 2, that there is no association between steroid dosage and behavior problems. We therefore propose that clinicians are likely interested in the effects of medications (or psychosocial interventions) on groups of children as well as on individuals, and both visual analysis and PTSA are useful techniques to address the respective concern.
Whether we choose to base conclusions from our results obtained via PTSA or visual techniques, repeated assessments helped illuminate the process of these children's adjustment over time. Specifically, eight (80%) of the children had CBCL scores in the nonclinical range at baseline (i.e., a prednisone-free data collection point at least 6 weeks prior to relapse and medication initiation). During relapse, five of eight children with normal baseline scores (mean T = 62.5) had CBCL scores above the 95th percentile for age and gender during at least one of the calling periods. Of these children, three had elevated scores on both CBCL subscales, three had elevated scores for aggressive behavior only, and one child had an elevated score for anxiety/depression only. The two children who had elevated scores at baseline also exhibited increased behavior problem scores during relapse. Again, the nature of the process of treatment-related behavioral change was detectable only by using this repeated measures methodology.
From a clinical perspective, findings such as these indicate specific risk periods for elevated behavioral problems that families and affected children need to know. As treatment regimens for disease relapses such as SSNS and other illnesses are often stretched out over several weeks or more, it is important to provide families anticipatory guidance related to potential steroid-induced effects. Families should receive additional support during periods when their children are at risk for side effects of significant concern. Appropriate anticipatory guidance could allow families to be better prepared for behavior problems both at home and at school.
From a research perspective, we acknowledge the possibility that, in
parents who report elevated behavior problems, their children's behavior might
have improved merely as an effect of support they perceived resulting from the
frequent phone calls required for participation in this study. Although the
investigators stressed that the purpose of the calls was for research only,
parents may have found that frequent reporting on their child's behavior
provided some relief for perceived difficulties. Other researchers have
reported intervention-related decreases in child behavior problems over the
course of daily phone calls (Chamberlain,
Moreland, & Reid, 1992). As a result, it has been suggested
that nondirective phone contact acts as a form of support for parents'
perceived behavioral symptoms in their children, especially for children with
high rates of behavior problems. Therefore, whether decreases in reported
behavior problems were linked solely to decreased medication dosage remains to
be determined.
In addition, although we did not tell parents we were collecting medication
dosage-related information, they likely determined this by the schedule of our
phone calls. As parents became aware of the correlation between phone call
timing and their children's medication schedule, they may have expected
medication-related changes. Such an expectancy phenomenon has been described
for other behavioral symptoms. In fact, the symptoms parents observed may have
been normal daily variations in their children's behavior. However, in
examining descriptive data on the entire group, approximately 30% of
participants reported little behavior change over the course of the phone
calls, thus raising doubt about a general response bias. Also, as discussed
earlier, previous research by Patterson
(1982) and others suggest that
daily reports are significantly less biased than parental reports asking
parents to aggregate perceptions over longer time periods such as weeks or
months. The forces behind parental perceptions of child behavior, whether
medication-based, psychosocial, or a combination of the two, remain to be
explored.
Limitations
Limitations of PTSA. Our goal is to provide a sound method for
gathering and analyzing data in small n situations, particularly to
examine psychosocial phenomena in rarely occurring pediatric chronic
conditions. The reader should be aware that even though PTSA offers
considerable advantages in analyzing data from small samples, like any other
analytic procedure, it has limitations. Primary among PTSA's limitations is
its capacity to handle nominal/ordinal individual-difference variables. Two
consistent concerns in pediatric psychology are age- and gender-related
effects. Unfortunately, PTSA cannot address effects of nominal variables. If
the researcher can collect data from a sufficient number of each gender, data
can be analyzed separately. If time-related effects are linear, follow-up
tests of regression coefficients can be conducted
(Cohen & Cohen, 1983).
Another concern is age-related effects. While the researcher can control for
age-related effects, significance tests of age-related effects can be
misleading, as their influence is overestimated by the percent variance
accounted for in the final equation
(Johnson, 1995).
More sophisticated procedures such as hierarchical linear modeling, latent growth curve analysis, and latent transition analysis can address nominal/individual difference variable effects. In addition, these modeling procedures allow for detection of systematic trends for missing data, which may occur, given the intensive nature of many repeated-measures designs applied to clinical populations. These procedures offer several advantages over PTSA, but they typically require sample sizes substantially larger than the one used in this study (or than are commonly available in single-site studies of rarely occurring pediatric chronic illnesses).
Limitations of This Study. The data used to demonstrate use of repeated measures (daily report) methodology and PTSA helped illustrate the major methodological concepts. However, as in any study, limitations of the results must be acknowledged. First, there is a potential confounding effect of illness symptoms on the outcomes (aggression and anxiety/depression). When children with nephrotic syndrome relapse, they are prescribed steroid medications presumably corresponding to symptom severity. Medication dosage is decreased corresponding to improvement in the child's condition. One could interpret the reduction in behavioral symptoms as relating to decreased illness severity rather than corresponding to decreased medication dosage, as we reported. Although it is unlikely that increased severity of illness caused behavioral changes of the magnitude reported in this study, it is possible that discomfort from disease-related symptoms such as edema may have contributed further to elevated behavior problems. One way to tease out the effects of medications versus disease symptomatology would be to assess a comparison sample of children taking medications other than highdose steroids. In some cases, children's disease symptoms do not respond to steroids (i.e., steroid-resistant), and after controlling for other illness variables, this population may constitute an appropriate comparison group.
Another limitation of this study was its sample size. As mentioned previously, effective sample size for PTSA is recommended to be somewhere between N and N x O, that is, somewhere between 10 and 200 for this study. Thus, our sample of 10 was at the lower limit of acceptability for PTSA.
|
Conclusions |
|---|
|
TopAbstractIntroductionMethodResultsDiscussionConclusionsReferences |
|---|
Within-subjects design with repeated temporal assessments was presented as a potential tool for examining pediatric psychology issues such as care-giving role strain, pediatric pain, treatment effects, and medication side effects. Moreover, these methods can help researchers exploit important clinical data related to low incidence disorders by, for example, maintaining the use of n = 1 methodology and its scientifically compelling aspect of replication. These methods also allow researchers to study naturally occurring clinical events or events that rely on timing and clinical judgment. Thus, it could even allow the pooling of data across settings, which would allow a more natural and rapid collection of data on clinical cases or events that have naturally low base-rates. Finally, results from the study example presented here provide applied evidence of the research and clinical utility of the procedures we described. Our goal is to make the methodology we presented accessible for researchers and clinicians to gain greater insight into pediatric psychology concerns.
|
Acknowledgments |
|---|
Portions of this study were supported by a Washington State University Dean's Office Minigrant. Dr. Moore was partially supported by Grants R01 MH 47458 and R01 MH 54257 from the Center of Studies of Violent Behavior and Traumatic Stress, National Institute of Mental Health (NIMH), and by Grant P50 MH 46690 from the Prevention Research Branch, NIMH. We thank families who participated in this study, Sheila Lee, and Dr. Wayne Osgood.
|
Notes |
|---|
1 In this study, we expected a linear relationship between the independent variable, steroid dosage, and the dependent variable, behavior scores, with behavior scores decreasing as a function of reduced medication dosage. However, the reader should be aware that procedures such as polynomial regression are available for capturing consistent but nonlinear relationships. These procedures also provide further control for systematic time-related trends. Due to the hypothesized linear relationship and limited sample size, nonlinear effects were not tested statistically in this study. Interested readers are referred to Kennard (1985
) for more information on
polynomial regression to address nonlinear effects. 
Received December 15, 1999; revision received May 1, 2000; revision received September 20, 2000; accepted October 30, 2000
|
References |
|---|
|
TopAbstractIntroductionMethodResultsDiscussionConclusionsReferences |
|---|
Achenbach, T. A. (1991). Manual for the Child Behavior Checklist/4-18 and 1991 Profile. Burlington, VT: University of Vermont Department of Psychiatry.
Achenbach, T. A. (1992). Manual for the Child Behavior Checklist/2-3 and 1992 Profile. Burlington, VT: University of Vermont Department of Psychiatry.
Allison, P. D. (1994). Using panel data to estimate the effects of events. Sociological Methods and Research, 23, 174-199.
Baer, D. M. (1977). "Perhaps it would be better not to know everything." Journal of Applied Behavior Analysis, 10, 176 -172.
Beans, B. (1999, November). Drug firms rely on psychologists' expertise. APA Monitor, pp. 1 , 14.
Bender, B. G., Lerner, J. A., & Poland J. E. (1991). Association between corticosteroids and psychologic change in hospitalized asthmatic children. Annals of Allergy, 61, 414 -419.
Bolger, N., Delongis, A., Kessler, R. C., & Schilling, E. A. (1989). Effects of daily stress on negative mood. Journal of Personality and Social Psychology, 57, 808-818.[Web of Science][Medline]
Chamberlain, P., Moreland, S., & Reid, K. (1992). Enhanced services and stipends for foster parents: Effects on retention rates and outcomes for children. Child Welfare, 71, 387-401.[Web of Science][Medline]
Cohen, J., & Cohen, P. (1983). Applied multiple regression/correlation analysis for the behavioral sciences. Hillsdale, NJ: Lawrence Erlbaum.
Cook, T. D., & Campbell, D. T. (1979). Quasiexperimentation: Design and analysis issues for field settings. Chicago, IL: Rand McNally College Publications.
Donaldson, G. W., & Moinpour, C. M. (1992). Strengthened estimates of individual pain trends in children following bone marrow transplantation. Pain, 42, 147-155.
Draper, N. R., & Smith. H. (1981). Applied regression analysis (2nd ed.). New York: John Wiley & Sons.
Engstroem, I., & Lindquist, B. L. (1998). Gastrointestinal disorders. In R. T. Ammerman & J. V. Campo (Eds.), Handbook of pediatric psychology and psychiatry, Vol. 2: Disease injury, and illness (pp. 206-223). Boston: Allyn & Bacon.
Greene, B. R., Blanchard, E. B., & Wan, C. K. (1994). Longterm monitoring of psychosocial stress and symptomatology in inflammatory bowel disease. Behavioral Research and Therapy, 32, 217 -226.
Harris, J. C., Carel, C. A., Rosenberg, L. A., Joshi, P., & Leventhal, B. G. (1986). Intermittent high dose corticosteroid treatment in childhood cancer: Behavioral and emotional consequences. Journal of the American Academy of Child and Adolescent Psychiatry, 25, 120 -124.
Hathaway, D. K., Winsett, R. P., Milstead, J., Wicks, M. N., & Gaber, A. O. (1996). Quality of life outcomes associated with variable posttransplant prednisone dosing regimens. Journal of Transplant Coordination, 6, 64 -68.
Hollingshead, A. B. (1975). Four factor index of social status. Unpublished manuscript, Yale University, New Haven.
Jaccard, J., & Wan, C. K. (1993). Statistical analysis of temporal data with many observations: Issues for behavioral medicine data. Annals of Behavioral Medicine, 15, 41-50.
Johnson, D. R. (1995). Alternative methods for the quantitative analysis of panel data in family research: Pooled time-series models. Journal of Marriage and the Family, 57, 1065 -1077.
Kennard, M. G. (1985). The use of higher-order polynomial coefficients as covariates in growth curve analysis. Biometrics, 41, 19 -28.[Medline]
Miller, D. J., Jelalian, E., & Stark, L. J. (1999). Cystic fibrosis. In A. J. Goreczny & M. Hersen (Eds.), Handbook of pediatric and adolescent health psychology (pp. 127-139). Boston: Allyn & Bacon.
Moore, K. J., Osgood, D. W., Larzelere, R. E., & Chamberlain, P. (1994). Use of pooled time series in the study of naturally occurring clinical events and problem behavior in a foster care setting. Journal of Clinical and Consulting Psychology, 4, 718-728.
Morrison, D. F. (1983). Applied linear statistical methods. Englewood Cliffs, NJ: Prentice-Hall.
Neter, J., Wasserman, W., & Kutner, M. H. (1985). Applied linear statistical models (2nd ed.). Homewood, IL: Irwin.
Ostrom, C. W., Jr. (1990). Time series analysis: Regression techniques (2nd ed., series no. 07-009). Sage University Paper series on Quantitative Applications in the Social Sciences. Newbury Park, CA: Sage.
Patterson, G. R. (1982). A social learning approach to family intervention: III. Coercive family process. Eugene, OR: Castalia.
Powers, S. W., Vannatta, K., Noll, R. B., Cool, V. A., & Stehbens, J. A. (1995). Leukemia and other childhood cancers. In M. C. Roberts (Ed.), Handbook of pediatric psychology (2nd ed., pp. 310-326). New York: Guilford.
Quittner, A. L., Opipari, L. C., Regoli, J. G., 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, 27, 275-290.
Richardson, G. M., & McGrath, P. J. (1983). Validity of the headache diary for children. Headache, 23, 184-187.[Web of Science][Medline]
Satel, S. (1990). Mental status changes in children
receiving glucocorticoids. Clinical Pediatrics, 29, 382-388.
Sayrs, L. W. (1989). Pooled time series analysis (series no. 07-070). Sage University Paper Series on Quantitative Applications in the Social Sciences. Newbury Park, CA: Sage.
SPSS, Inc. (1994). SPSS 6.1 syntax reference guide. Chicago: Author.
Stevens, J. (1992). Applied multivariate statistics for the social sciences (2nd ed.). Hillsdale, NJ: Lawrence Erlbaum.
Warshaw, B. L. (1994). Nephrotic syndrome in children. Pediatric Annals, 23, 495 -503.[Web of Science][Medline]
Werle, M. A., Murphy, T. B., & Budd, K. S. (1993). Treating chronic food refusal in young children: Home-based parent training. Journal of Applied Behavior Analysis, 26, 421-433.[Web of Science][Medline]
West, S. G., & Hepworth, S. G. (1991). Statistical issues in the study of temporal data: Daily experiences. Journal of Personality, 59, 609 -662.[Web of Science][Medline]
CiteULike 
Connotea 
Del.icio.us What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||