Valuation of Child Behavioral Problems from the Perspective of US Adults

Abstract

Objective. To assess preferences between child behavioral problems and estimate their value on a quality-adjusted life year (QALY) scale. Methods. Respondents, age 18 or older, drawn from a nationally representative panel between August 2012 and February 2013 completed a series of paired comparisons, each involving a choice between 2 different behavioral problems described using the Behavioral Problems Index (BPI), a 28-item instrument with 6 domains (Anxious/Depressed, Headstrong, Hyperactive, Immature Dependency, Anti-social, and Peer Conflict/Social Withdrawal). Each behavioral problem lasted 1 or 2 years for an unnamed child, age 7 or 10 years, with no suggested relationship to the respondent. Generalized linear model analyses estimated the value of each problem on a QALY scale, considering its duration and the child’s age. Results. Among 5207 eligible respondents, 4155 (80%) completed all questions. Across the 6 domains, problems relating to antisocial behavior were the least preferred, particularly the items related to cheating, lying, bullying, and cruelty to others. Conclusions. The findings are the first to produce a preference-based summary measure of child behavioral problems on a QALY scale. The results may inform both clinical practice and resource allocation decisions by enhancing our understanding of difficult tradeoffs in how adults view child behavioral problems. Understanding US values also promotes national health surveillance by complementing conventional measures of surveillance, survival, and diagnoses.

Keywords

Behavioral Problems Index quality-adjusted life years QALY discrete choice experiments patient-reported outcomes

Introduction

A year of life with no health problems is one quality-adjusted life year (QALY), and its value serves as a common preference-based metric in comparative effectiveness research (CER). Specifically, any episode with health problems may be summarized by its equivalence in reduced lifespan with no health problems (i.e., QALYs) using responses from a health valuation survey. For example, a participant might be asked, “Which do you prefer: a year in mild pain or a 6-month loss in lifespan with no health problems?” Responses to such questions quantify preferences between health outcomes without referencing other considerations (e.g., money). The purpose of this study is to assess preferences between child behavioral problems and estimate their value on a QALY scale.

Although multiple studies have assessed preferences between child health scenarios using responses from valuation surveys,^1–5 their results are typically not linked to any child health instrument, and therefore cannot directly summarize child health as measured in CER data. Furthermore, child health valuation studies must include a description of lifespan or risk of death in their scenarios to estimate the value of problems on a QALY scale. For example, an Australian study incorporated health scenarios as described by the Child Health Utility 9D, but excluded a lifespan attribute (i.e., no QALYs).⁶ To date, only 2 studies are based on a child health–related quality-of-life (HRQoL) instrument and estimate loss on a QALY scale. Both studies valued the Pediatric Asthma Health Outcome Measure (PAHOM) using a standard gamble task, with samples of adults from Seattle, Washington (N = 94), and Birmingham, Alabama (N = 261).^7,8 Due to this paucity in the literature, CER studies using the Health Utilities Index⁹ and the EQ-5D⁴ have applied the same QALY weights to both child and adult outcomes, as if their experiences were interchangeable. However, we know from other literature⁵ that adults often express preferences about health care differently for children than for adults, especially when resources are limited.

The persistent lack of data on preferences for child health outcomes must be addressed. A PubMed search for the terms “children health-related quality of life” identified 3343 articles, with only 127 of them published prior to 2000. The passing of the 2010 US Patient Protection and Affordable Care Act and the formation of the Patient-Centered Outcomes Research Institute (PCORI) have strengthened the importance of HRQoL as a patient-centered outcome.^10,11 Furthermore, our expanding technological capacity to systematically collect real-time data has the potential to revolutionize our understanding of health-related experiences among children for CER and public health surveillance. In summary, clinical guidelines, resource allocations, and policy decisions are informed by formally weighing evidence on child health outcomes; yet, only one pediatric instrument (i.e., PAHOM) has been directly summarized on a QALY scale. In this study, we take the perspective of US adults and assess preferences between child health behavioral problems as described by the Behavioral Problems Index (BPI).

Developed by James L. Peterson and Nicholas Zill¹² and based on earlier work by Thomas Achenbach,¹³ the BPI is a seminal measure of child behavior reported by mothers using 28 3-level items along 6 domains: Anxious/Depressed, Headstrong, Hyperactive, Immature Dependency, Anti-social, and Peer Conflict/Social Withdrawal.^14–16 This validated instrument has been used broadly in a multitude of surveys, including the Panel Survey of Income Dynamics,¹⁷ the Chicago School Readiness Project,¹⁸ the Child Health Supplements to the National Health Interview Survey,^19

–22 and the National Longitudinal Survey of Youth.^23

–28 The BPI clearly covers important child HRQoL domains, but does not capture all domains of child health, or the impact of behavioral problems on the HRQoL of others (e.g., caregivers). It is also not a “generic” measure of broader aspects of HRQoL, such as the Health Utilities Index⁹ or the EQ-5D.⁴ By summarizing the BPI on a QALY scale, this study provides a new tool that enhances the potential contribution of existing datasets for CER.

Methods

Participants

To inform medical decision making and health policy, CER requires measurement and valuation.²⁹ Measurement typically involves surveys of health outcomes completed by patients (e.g., children) or their proxies (e.g., parents or caregivers). Valuation requires surveys of preferences from the perspective of decision makers (e.g., general population). For this valuation study, we surveyed adults (instead of children), age 18 years or older, who resided in the United States, because adults typically make health care decisions for children. Respondents were recruited from a preexisting nationally representative panel of US adults. To promote concordance with the 2010 US Census, we used 18 demographic subgroups (all combinations of 2 genders, 3 age groups, and 3 race/ethnicity groups). Once a minimum number of respondents for a subgroup were received, additional respondents belonging to that subgroup were not allowed (or paid) to complete the survey. The survey was administered online between August 7, 2012, and February 5, 2013. The protocol, including its sampling design and survey instrument, was adapted from the PROMIS-29 valuation study (1R01CA160104)³⁰ and approved by the University of South Florida Institutional Review Board (USF IRB #8236).

Survey

After consenting, respondents completed a screener in which they reported their current US state of residence, zip code, birthdate, race, Hispanic ethnicity, educational attainment, and household income (Table 1). After the screener, respondents proceeded to the survey, which was composed of health, paired-comparison, and follow-up components. The health component included the PROMIS-29, a validated measure supported by a National Institutes of Health initiative as a measure of adult HRQoL.³¹ The follow-up component asked about the respondent’s experience with parenting and selected childhood health conditions, and offered an opportunity to leave survey feedback.

Table 1

Respondent Characteristics by Completion and Compared to the 2010 US Census*

	Dropout	Terminated	Completed		Completed
	N = 805 # (%)	N = 247 # (%)	N = 4155 # (%)	P-value	N = 4155 %	US 2010 Census%
Age in Years
18–34	192 (13.67)	77 (5.48)	1136 (80.85)	0.017	27.34	30.58
35–54	271 (14.74)	87 (4.73)	1481 (80.53)		35.64	36.70
55 and older	342 (17.42)	83 (4.23)	1538 (78.35)		37.02	32.72
Sex
Male	397 (15.67)	123 (4.86)	2013 (79.47)	0.843	48.45	48.53
Female	408 (15.26)	124 (4.64)	2142 (80.10)		51.55	51.47
Race
White	651 (15.37)	190 (4.49)	3394 (80.14)	0.337	81.68	74.66
Black or African American	105 (16.38)	43 (6.71)	493 (76.91)		11.87	11.97
American Indian or Alaska Native	5 (17.86)	0 (0.00)	23 (82.14)		0.55	0.87
Asian	21 (14.69)	8 (5.59)	114 (79.72)		2.74	4.87
Native Hawaiian or Pacific Islander	10 (20.00)	1 (2.00)	39 (78.00)		0.94	0.16
Some other race	—	—	—		—	5.39
Two or more races	13 (11.82)	5 (4.55)	92 (83.64)		2.21	2.06
Hispanic Ethnicity
Hispanic or Latino	97 (15.80)	33 (5.37)	484 (78.83)	0.699	11.65	14.22
Not Hispanic or Latino	708 (15.41)	214 (4.66)	3671 (79.93)		88.35	85.78
Educational Attainment Among Age 25 or Older
Less than high school	86 (17.17)	31 (6.19)	384 (76.65)	0.001	9.24	14.42
High school graduate	211 (18.89)	59 (5.28)	847 (75.83)		20.39	28.50
Some college, no degree	147 (15.30)	54 (5.62)	760 (79.08)		18.29	21.28
Associate’s degree	66 (13.41)	14 (2.85)	412 (83.74)		9.92	7.61
Bachelor’s degree	216 (13.76)	60 (3.82)	1294 (82.42)		31.14	17.74
Graduate or professional degree	22 (12.87)	6 (3.51)	143 (83.63)		3.44	10.44
Refused/don’t know	4 (26.67)	0 (0.00)	11 (73.33)		0.26	—
Household Income
$14,999 or less	77 (14.37)	29 (5.41)	430 (80.22)	0.129	10.35	13.46
$15,000–$24,999	123 (16.40)	36 (4.80)	591 (78.80)		14.22	11.49
$25,000–$34,999	105 (15.49)	37 (5.46)	536 (79.06)		12.90	10.76
$35,000–$49,999	146 (15.72)	35 (3.77)	748 (80.52)		18.00	14.24
$50,000–$74,999	92 (12.87)	37 (5.17)	586 (81.96)		14.10	18.28
$75,000–$99,999	86 (18.30)	11 (2.34)	373 (79.36)		8.98	11.81
$100,000–$149,999	49 (12.50)	21 (5.36)	322 (82.14)		7.75	11.82
$150,000 or more	44 (16.54)	16 (6.02)	206 (77.44)		4.96	8.14
Refused/don’t know	83 (17.62)	25 (5.31)	363 (77.07)		8.74	—

Age, sex, race, and ethnicity estimates for the United States are based on 2010 Census Summary File 1. Educational attainment and household income are based on 2010 American Community Survey 1-Year Estimates. Unlike the US Census, the American Community Survey excluded adults not in the community (e.g., institutionalized) and describes income by the proportion of households, not of adults.

Dropout refers to a respondent who exited the study either voluntarily or due to a loss of internet connection. Terminated refers to a respondent who was removed from the survey due to technical requirements (e.g., JavaScript not enabled).

Paired-Comparison Component

A paired comparison is a choice-based question that asks a respondent about his or her preference between 2 alternatives (e.g., Coke vs. Pepsi). Responses show how choices change with different combinations of alternatives. Each respondent first completed 3 example paired-comparison questions: “Which do you prefer?” (Apple, Orange), (Good Health, Poor Health), or (Bad Health, Poor Health). The “Bad Health” versus “Poor Health” question was included to prepare respondents for potentially more challenging descriptions later on in the survey. Next, respondents were randomly assigned a base scenario (a scenario used as a reference point that does not change) and asked to complete a series of paired comparisons building from this base scenario. The base scenario described the age of an unnamed child (7 or 10 years old) and the duration of a behavioral problem (1 or 2 years). Aside from the 3 examples, the number of paired comparisons ranged from 33 to 40 pairs. Due to space considerations, the appendix provides didactic descriptions of the pairs, adjectival statements, paired selection, econometrics, and results, which are summarized in this section.

Initial pairs asked respondents to choose between a behavioral problem and a loss in lifespan given the assigned base scenario. For example, the paired comparison in Figure 1 has a base scenario for a 7-year-old child. In this task, the respondents must choose between a 5-year reduction in child lifespan (i.e., loss of 5 QALYs) and a behavioral problem, specifically, problems relating to Anxiety/Depression, Headstrong, and Immature Dependency for 1 year. For this 7-year-old child, the left alternative represents a 6-year lifespan with no problems, and the right is an 11-year lifespan that starts with 1 year of behavioral problems. For these initial pairs, the life span extended 10 years after the behavioral problem ended, which follows common practice in time tradeoff (TTO) tasks and allows for sufficient range in loss of lifespan.³² Remaining pairs asked respondents to choose between 2 behavioral problems. Respondents were assigned to see problems, described using statements derived from the BPI, for just 1 of 2 durations (1 year or 2 years). The use of 2 durations was included to assess the constant proportionality assumption. The episodic random utility model (ERUM) assigns value based on episode attributes, including duration (relaxing the constant proportionality assumption), whereas a health state approach (aka the instant random utility model [IRUM]) assumes constant proportionality (i.e., 2 years of problems have twice the value of 1 year).^33
–35 All pair sequences were randomly ordered to reduce sequence effects. Pairs were assigned to respondents according to 18 demographic subgroups to strengthen concordance with the 2010 US Census at the pair level.

Statistical Analysis

Screener responses of those who dropped out, were terminated, and completed the survey were compared using chi-square tests and shown alongside the 2010 US Census results (Table 1). Responses to the 211 pairs were stratified by 4 base scenarios (i.e., each respondent sees only 1 of the 4 base scenarios). These pairs (shown in the appendix) include 136 item pairs asking for the respondent’s preference between item-specific attributes within each domain, 12 domain pairs asking about preference between domain-specific attributes, 15 class pairs asking about preference between groups of domains, and 48 pairs asking about preference between behavioral problems and losses in lifespan (Figure 1). Figure 2 illustrates responses to 48 pairs comparing behavioral problems and losses in lifespan; however, all pair results are included in the appendix.

The BPI has 28 3-level items and captures up to 56 problems (i.e., differences in the ordinal scale; see Tables 2 and 3). For each of the 844 pairs (4 base scenarios × 211 pairs; see the appendix), the sample probability of choosing A over B, p_k, was approximately normally distributed by the central limit theorem and served as the dependent variable of a generalized linear model. Each alternative (A and B) was represented by a linear regression, d_h, that includes 56 coefficients, one for each indicator variable of a difference in ordinal scale. For example, the first item in the Anxiety/Depression (AD) domain is sudden change in mood or feeling, and the value of the step from no to sometimes is represented by AD_1,1, such that the first subscript represents the item and the second subscript represents the lower level of the difference in the ordinal scale. The coefficients of the generalized linear model (GLM) were estimated by minimizing the weighted sum of squared error, ${\sum_{k = 1}^{844} (P (A_{K} > B_{K}) - p_{k})}^{2} / σ_{k}^{2}$ , where P(A_k>B_k) = d_B / ( d_A+d_B) and σ²_k = p_k × (1−p_k) / n_k.^36,37 To understand the motivation for the GLM approach, it is important to recognize that given a pair sample size, N_k, and a population probability, P_k, its sample probability, p_k, is approximately normally distributed if N_k × P_k > 5 or N_k × (1−P_k) > 5. In this study, each pair had 50 or more responses; therefore, each of the 844 sample probabilities are approximately normally distributed, assuming that their P_k are between 0.1 and 0.9. In the GLM, P(A_k > B_k) is a link function that best fits these normally distributed data. Furthermore, the model was re-estimated after stratifying the pairs by base scenario. The significance level was set at 0.05, and 95% bootstrapped confidence intervals were computed for all parameters.

Table 2

The Value of Anxiety/Depression, Headstrong, and Hyperactive Problems

	None Vs. Sometimes	Sometimes Vs. Often	None Vs. Often	Domain
Child Behavioral Problem by Domain and Item	QALY (CI)	QALY (CI)	QALY	QALY
Anxiety/Depression
Sudden changes in mood or feeling	0.034	0.075	0.109
	(0.025–0.041)	(0.060–0.090)
Feels or complains that no one loves him or her	0.04	0.103	0.143
	(0.03–0.049)	(0.085–0.128)
Too fearful or anxious	0.031	0.082	0.113
	(0.023–0.037)	(0.067–0.099)
Feels worthless or inferior	0.084	0.254	0.338
	(0.069–0.104)	(0.231–0.320)
Unhappy, sad, or depressed	0.051	0.239	0.290	0.993
	(0.039–0.062)	(0.216–0.325)
Headstrong
Rather high-strung, tense, and nervous	0.046	0.096	0.142
	(0.037–0.053)	(0.081–0.118)
Argues too much	0.02	0.061	0.081
	(0.014–0.023)	(0.051–0.071)
Disobedient at home	0.025	0.183	0.208
	(0.018–0.029)	(0.161–0.23)
Stubborn, sullen, and irritable	0.031	0.090	0.121
	(0.024–0.036)	(0.074–0.107)
A very strong temper and sometimes loses it easily	0.102	0.230	0.332	0.884
	(0.085–0.129)	(0.208–0.295)
Hyperactive
Difficulty concentrating or paying attention for long	0.064	0.128	0.192
	(0.050–0.075)	(0.111–0.144)
Easily confused or sometimes seems to be in a fog	0.146	0.243	0.389
	(0.123–0.168)	(0.215–0.287)
Impulsive or sometimes acts without thinking	0.067	0.211	0.278
	(0.053–0.077)	(0.187–0.242)
A lot of difficulty getting his or her mind off certain thoughts	0.122	0.196	0.318
	(0.104–0.139)	(0.171–0.23)
Restless, sometimes overly active, and sometimes cannot sit still	0.069	0.154	0.223	1.400
	(0.056–0.079)	(0.134–0.176)

CI, Confidence interval; QALY, quality-adjusted life year.

Table 3

The Value of Immature Dependency, Anti-social, and Peer Conflict/Social Withdrawal Problems

	None Vs. Sometimes	Sometimes Vs. Often	None Vs. Often	Domain
Child Behavioral Problem by Domain and Item	QALY (CI)	QALY (CI)	QALY	QALY
Immature Dependency
Clings to adults	0.014	0.051	0.065
	(0.009–0.016)	(0.040–0.059)
Cries too much	0.040	0.096	0.136
	(0.031–0.047)	(0.077–0.119)
Demands a lot of attention	0.021	0.067	0.088
	(0.016–0.025)	(0.055–0.079)
Too dependent on others	0.034	0.083	0.117	0.406
	(0.026–0.040)	(0.066–0.102)
Anti-social
Cheats or sometimes tells lies	0.448	1.026	1.474
	(0.390–0.659)	(0.943–1.697)
Bullies others; cruel and mean to others	1.191	1.628	2.819
	(1.122–1.881)	(1.555–2.917)
Not sorry after misbehaving	0.267	0.314	0.581
	(0.214–0.375)	(0.273–0.447)
Breaks things on purpose; sometimes deliberately destroys things	0.470	0.550	1.020
	(0.406–0.699)	(0.474–0.879)
Disobedient at school	0.123	0.411	0.534
	(0.093–0.171)	(0.351–0.623)
Trouble getting along with teachers	0.142	0.260	0.402	6.830
	(0.110–0.197)	(0.211–0.393)
Peer Conflict/Social Withdrawal
Trouble getting along with other children	0.301	0.438	0.739
	(0.205–0.463)	(0.331–0.770)
Not liked by other children	0.246	0.469	0.715
	(0.164–0.376)	(0.352–0.816)
Withdrawn and sometimes does not get involved with others	0.299	0.307	0.606	2.060
	(0.201–0.452)	(0.229–0.541)

CI, Confidence interval; QALY, quality-adjusted life year.

Results

Survey Participation

Of the 11,496 respondents recruited for this study, 1075 (9.35%) visited only the consent page, 190 (1.65%) reported nonconsent, and 408 (3.55%) dropped out during the screener. Among the 9823 respondents who completed the screener, 2669 (27.17%) belonged to a subgroup that already had sufficient respondents, and 1947 (19.82%) failed the screener requirements. Among the 5207 respondents who were allowed to enter the survey, 805 (15.46%) dropped out during the survey, and 247 (4.74%) were terminated due to technical requirements (e.g., JavaScript not enabled). The 4155 respondents who completed the survey were younger than those who dropped out, older than those who were terminated, and better educated than those who did not complete the survey (see Table 1). Compared to the 2010 US Census, the analytical sample was demographically similar yet better educated, with small differences at the extremes in annual household incomes (less than $15,000 and greater than $150,000). The sample sizes of the 844 paired comparisons pertaining to the BPI ranged from 51 to 80 respondents. The median survey duration was 25.21 minutes (interquartile range, 19.5–34.2 minutes). Most participants who completed the survey reported that the survey was easy to understand (71%) and navigate (87%).

Choices

Figure 2 illustrates the percentage of respondents who preferred reduced lifespan over additional behavioral problems, combining responses across the 4 base scenarios. The value of each behavioral problem on a QALY scale is defined by where its line crosses 50% on the y-axis, because this is the point where exactly half of respondents prefer reduced lifespan over the behavioral problem. As expected, the percentages form lines that are largely parallel and decreasing. The height of a line implies greater willingness to sacrifice lifespan (i.e., in place of a less desirable problem). The topmost line (in this case, denoted with white triangles) indicates the least desirable problems compared to the lines below it. The third line (white squares) is higher (less desirable) than the fourth line (black squares) by construction, because it represents a difference between none and often compared to the difference between sometimes and often in the Hyperactivity domain. Interestingly, problems relating to Anxiety/Depression, Headstrong, and Immature Dependency (triangle lines) were less desirable than problems relating to Hyperactivity (square lines).

Value of Child Behavioral Problems

Tables 2 and 3 describe the value of child behavioral problems on a QALY scale assuming that each problem lasts for 1 year. For all items, the difference between levels 1 (never) and 2 (sometimes) was less than the difference between levels 2 (sometimes) and 3 (often). The value of an item is the sum of its differences in levels, and the value of a domain is the sum of its item values. For example, consider the first item within Hyperactivity: difficulty concentrating or paying attention for long. The value of having this problem sometimes for 1 year equals a loss of 0.064 QALYs, and the value of the difference between having the problem sometimes versus often equals a loss of 0.128 QALYs; therefore, the value of the item is 0.192 (i.e., 0.064+0.128). The value of the domain Hyperactivity is 1.400 QALYS (i.e., 0.192+0.389+0.278+0.318+0.223), which includes the values of its 5 items.

Across the 6 domains, problems relating to Anti-social behavior were by far the least desirable, particularly the items relating to cheating, lying, bullying, and cruelty to others. This domain was less desirable than all other domains combined (6.630 vs. 5.743 QALYs), possibly because it imposes a burden on others by definition. The least undesirable domain was Immature Dependency (0.406 QALYs), which had the least undesirable item, sometimes clings to adults (0.014 QALYs).

Differences by Age and Problem Duration

After stratifying the estimation by base scenario (see the appendix), we found that only 2 of the 56 problems had significant differences in value by child age at the 1-year and 2-year durations. Both differences relate to the Hyperactivity item, restless, overly active, and cannot sit still. The first shows a higher value (i.e., less preferable) on sometimes hyperactive at age 10 compared to age 7 years (0.057 difference in QALYs; 95% CI, 0.014–0.101). The second places a lower value (i.e., more preferable) on often hyperactive at age 10 compared to age 7 (0.044 difference in QALYs; 95% CI, 0.001–0.096). In addition, we found many significant differences by problem duration at both 7 and 10 years of age (29 and 12 out of 56, respectively). The signs and magnitude of the duration effects suggest that doubling the duration of a behavioral problem less than doubles its value.

Discussion

Given the wealth of available data using the BPI, the importance of CER for medical decision making and health policy, and our expanding technological capacity to systematically collect health data, these results have the potential to greatly improve our understanding of child behavioral problems from the perspective of US adults.

This study found that half of US adults would prefer a 7-year reduction in a child’s dozen-year lifespan than for there to be 1 year of that child’s antisocial behavior (i.e., often cheats, lies, bullies, and breaks things on purpose, and not sorry after misbehaving; often has trouble getting along with teachers; and sometimes is disobedient at school). This finding may reflect the readily observable impact of antisocial behavior on the child as well as the impacts experienced by his or her family, school, and community. This 7:1 ratio is clearly quite large, but it emphasizes the importance of access to child mental health services and funding of pediatric surveillance and research. Keeping in mind that the losses in lifespan occur in early adulthood and the suffering occurs while a child, the prevention of bullying among children may be worth the seemingly high loss in adult lifespan from the perspective of US adults, possibly due to the perceived long-term and communal consequences of antisocial behavior or the fact that adults have had their “fair innings” (i.e., everyone is entitled to some “normal” span of health, anyone failing to achieve this has been cheated, and anyone getting more than this is “living on borrowed time”).^32,38,39

On the methodological side, our study explores 2 key assumptions underlying QALYs: age independence and constant proportionality in time.³³ The former implies that adults’ preferences regarding child behavioral problems do not depend on the age of the child experiencing it, and the latter states that the value of a problem is in a constant proportion to its duration. For the first to hold true, the preferences concerning losses experienced by a 7-year-old would be the same as those experienced by a 10-year-old. For the second, our estimate of the loss in QALYs caused by a problem lasting 2 years should be twice as large as the loss due to the same problem lasting 1 year. This assumption is rarely tested among adults, although it is known to be violated in previous studies.⁴⁰ For children, we hypothesized that both assumptions may be violated because health norms vary greatly between younger and older children (e.g., self-care). Future studies should further investigate these issues and how they may differ between child and adult populations.

Our findings show violations of both assumptions; however, the differences between ages 7 and 10 seem minor compared to the significant violation of constant proportionality: The loss in QALYs for having 2 years of behavioral problems was considerably less than twice the loss for having 1 year of problems. This violation of constant proportionality may be due to faith in the resilience of children; their families, schools, and communities; and a child’s ability to adapt to behavioral problems with time. This study, like most valuation studies, asked the respondent to assume that children can be completely rehabilitated after a 1- or 2-year period; however, the respondent’s failure to do so (i.e., once a bad egg, always a bad egg) may lead to overestimation in the first year. We recommend further research in this area. It should be noted that constant proportionality is an assumption of the QALY metric, not of the methods in this analysis.

To date, no study has assessed the value of child behavioral problems on a QALY scale. In the tradition of adult health valuation, paired comparisons have been used for the valuation of SF-36, PROMIS-29, and EQ-5D instruments.^41
–43 Pairs including a 10-year lifespan are used frequently in the TTO task, common in adult health valuation, which uses an adaptive series of paired comparisons.^15,16,44,45 Unlike TTO-based studies, this preference study is based solely on nonadaptive paired comparisons. Simplifying the comparisons and task reduces cognitive burden, expedites response, and reduces the use of some simplifying heuristics that may introduce bias in complex choices.⁴⁶ Future research will show whether the results of this approach might replace the traditional adaptive tasks (e.g., standard gamble).

The study design also raises questions about the effects of using unnamed children and 10-year time horizons in the valuation of child health. Usually, adult participants are asked to value their own health, not the health of an unnamed child. The results might differ with a familial or named description of the child (e.g., by gender, socioeconomic status, or disease), particularly among parents or survivors of childhood diseases. Also, the preferences may change if the reductions in lifespan occurred later in the child’s life (as an older adult), instead of 10 years later when the child becomes a young adult. Although 10 years is the most common time horizon, a wide range of lifespans have been incorporated in TTO studies.⁴⁷ The use of the 10-year time horizon may be particularly problematic when applied to the valuation of child health. For example, the child in Figure 1 has a tragically short lifespan regardless of the choice (13 and 18 years, respectively), which may confound the preference elicitation task. However, a longer horizon would likely increase a willingness to reduce adult lifespan (i.e., intertemporal discounting) to prevent child behavioral problems, further inflating the seemingly high estimates (i.e., the 7:1 ratio). The issues of unnamed children, “fair innings,” and the 10-year time horizon have been raised in other health valuation literature, and they represent an unresolved area for which there is no standard.^32,38 The present article adds to the sparse literature on adult valuations of child health outcomes, but its novelty also suggests that the results should be approached with caution. Throughout time, as this literature grows, these issues and how they affect valuation may be tested further.

A typical use of study findings like ours would be to facilitate health technology assessment (HTA) or health economic evaluation of the costs and benefits of alternative medical interventions and technologies. This study, like most valuation studies, used an additive regression model on a QALY scale in which often problems comprise the sum of the difference between none and sometimes and the difference between sometimes and often. Although HTA and economic evaluation remain controversial, especially in the United States,⁴⁸ they provide an important framework for medical decision making at both the clinical and managerial levels.^5,49
–51 For child health, one major shortcoming of HTA to date has been the lack of QALY weights for HRQoL instruments that are specific to children.⁵² Adult HRQoL instruments often focus on aspects of health that are more relevant to the elderly and miss many important characteristics for children. Moreover, it is inappropriate to assume that how adults experience HRQoL is the same as how children experience it.⁹ Future studies may apply our results to existing datasets or examine the preferences of adult subsamples (e.g., parents and caregivers) and children themselves.

Our study helps to improve quantitative and economic evaluation in medical decision making and HTA by providing a preference-based measure of child behavioral problems. Given the growing emphasis on patient-centered outcomes,⁵³ this advancement is relevant. Our results also show that future HTA for child health policy will need to consider methodological improvements because we found violations of both the constant proportionality in duration and age independence assumptions.

Footnotes

Acknowledgements

The authors thank the staff in Dr. Craig’s lab at Moffitt Cancer Center for their contributions to the research and creation of this article: Michelle Owens, MA (psychology; research coordinating); Catherine Blackburn, BS (literature review); and Carol Templeton, BA (technical writing; copyediting).

There are no conflicts of interest.

Financial support for this research was provided by a grant from the National Institutes of Health, Department of Health and Human Services, through the National Cancer Institute (1R01CA160104) and by Dr. Craig’s support account at Moffitt Cancer Center. The funding agreements ensured the authors’ independence in designing the study, interpreting the data, writing, and publishing the report.

Supplementary material for this article is available on the Medical Decision Making Web site at .

References

Carroll

Downs

. Improving decision analyses: parent preferences (utility values) for pediatric health outcomes. J Pediat. 2009;155(1):21–5.

Friedman

Reed

Weinfurt

Kahler

Walter

Schulman

. Parents’ reported preference scores for childhood atopic dermatitis disease states. BMC Pediatr. 2004;4(1):21.

Hauber

Itzler

Johnson

. Healthy-days time equivalents for outcomes of acute rotavirus infections. Vaccine. 2011;29(45):8086–93.

Secnik

Matza

Cottrell

Edgell

Tilden

Mannix

. Health state utilities for childhood attention-deficit/hyperactivity-disorder based on parent preferences in the United Kingdom. Med Decis Making. 2005;25(1):56–70.

Ungar

. Economic Evaluation in Child Health. Oxford: Oxford University Press; 2010.

Ratcliffe

Couzner

Flynn

. Valuing child health utility 9D health states with a young adolescent sample: a feasibility study to compare best-worst scaling discrete-choice experiment, standard gamble and time trade-off methods. Appl Health Econ Health Policy. 2011;9(1):15–27.

Chiou

Weaver

Bell

Lee

Krieger

. Development of the multi-attribute Pediatric Asthma Health Outcome Measure (PAHOM). Intl J Qual Health Care. 2005;17(1):23–30.

Gerald

McClure

Harrington

Moore

Hernandez-Martinez

Gerald

. Measurement characteristics of the Pediatric Asthma Health Outcome Measure. J Asthma. 2012;49(3):260–6.

Petrou

Kupek

. Estimating preference-based Health Utilities Index Mark 3 utility scores for childhood conditions in England and Scotland. Med Decis Making. 2009;29(3):291–303.

10.

Patient Protection and Affordable Care Act. Pub. L. 111-148, 124 stat. 119 (as amended by the Health and Education Reconciliation Act, pub. L. 111-152, 124 stat. 1029, subtitle d: patient centered outcomes research). 2010.

11.

Selby

Beal

Frank

. The Patient-Centered Outcomes Research Institute (PCORI) national priorities for research and initial research agenda. JAMA. 2012;307(15):1583–4.

12.

Peterson

Zill

. Marital disruption, parent–child relationships, and behavior problems in children. J Marr Fam. 1986;48(2):295–307.

13.

Achenbach

Edelbrock

. Behavioral problems and competencies reported by parents of normal and disturbed children aged 4 through 16. Mono Soc Res Child Develop. 1981;46(1):1–82.

14.

Bureau of Labor Statistics. National longitudinal survey of youth 1979 children and young adults. Washington, DC: Bureau of Labor Statistics; 2013. https://www.nlsinfo.org/content/cohorts/nlsy79-children/topical-guide/assessments/behavior-problems-index-bpi. Last accessed June 29, 2015.

15.

Ravens-Sieberer

Wille

Badia

. Feasibility, reliability, and validity of the EQ-5D-Y: results from a multinational study. Qual Life Res. 2010;19(6):887–97.

16.

Wille

Badia

Bonsel

. Development of the EQ-5D-Y: a child-friendly version of the EQ-5D. Qual Life Res. 2010;19(6):875–86.

17.

Christakis

Zimmerman

. Violent television viewing during preschool is associated with antisocial behavior during school age. Pediatrics. 2007;120(5):993–9.

18.

Raver

Jones

Li-Grining

Metzger

Champion

Sardine

. Improving preschool classroom processes: preliminary findings from a randomized trial implemented in head start settings. Early Child Res Quart. 2008;23(1):10–26.

19.

Bussing

Halfon

Benjamin

Wells

. Prevalence of behavior problems in US children with asthma. Arch Pediat Adoles Med. 1995;149(5):565–72.

20.

Byrd

Weitzman

Auinger

. Increased behavior problems associated with delayed school entry and delayed school progress. Pediatrics. 1997;100(4):654–61.

21.

Byrd

Weitzman

Lanphear

Auinger

. Bed-wetting in us children: epidemiology and related behavior problems. Pediatrics. 1996;98(3):414–9.

22.

Gortmaker

Walker

Weitzman

Sobol

. Chronic conditions, socioeconomic risks, and behavioral-problems in children and adolescents. Pediatrics. 1990;85(3):267–76.

23.

Christakis

Zimmerman

DiGiuseppe

McCarty

. Early television exposure and subsequent attentional problems in children. Pediatrics. 2004;113(4):708–13.

24.

Crockett

Eggebeen

Hawkins

. Fathers presence and young childrens behavioral and cognitive adjustment. J Fam Issues. 1993;14(3):355–77.

25.

King

. Variation in the consequences of nonresident father involvement for childrens well-being. J Marriage Fam. 1994;56(4):963–72.

26.

King

. Nonresident father involvement and child well-being: can dads make a difference? J Fam Issues. 1994;15(1):78–96.

27.

Lumeng

Gannon

Cabral

Frank

Zuckerman

. Association between clinically meaningful behavior problems and overweight in children. Pediatrics. 2003;112(5):1138–45.

28.

Weitzman

Gortmaker

Sobol

. Maternal smoking and behavior problems of children. Pediatrics. 1992;90(3):342–9.

29.

Craig

Reeve

. Patient-reported outcomes and preference research: Igniting the candle at both ends and the middle. ISPOR Conn. 2012 Sep–Oct:18(5):24.

30.

Craig

Reeve

. Methods report on the PROMIS valuation study: year 1. Moffitt Cancer Center. http://labpages.moffitt.org/craigb/Publications/P29VYr2_140328.pdf

31.

PROMIS-29 profile v1.0. 2008–2012. https://www.assessmentcenter.net/ac1//files/pdf/44b7636201a34267a9213db7f69f2c6d.pdf. Last accessed June 29, 2015.

32.

Feeny

. Standardization and regulatory guidelines may inhibit science and reduce the usefulness of analyses based on the application of preference-based measures for policy decisions. Med Decis Making. 2013;33(3):316–9.

33.

Craig

Busschbach

. The episodic random utility model unifies time trade-off and discrete choice approaches in health state valuation. Popul Health Metr. 2009;7:3.

34.

Craig

Busschbach

Salomon

. International valuation set for EQ-5D health states. Value Health. 2008;11(3):A177–A.

35.

Craig

Busschbach

Salomon

. Modeling ranking, time trade-off, and visual analog scale values for EQ-5D health states: a review and comparison of methods. Med Care. 2009;47(6):634–41.

36.

McFadden

. Conditional logit analysis of qualitative choice behavior. In: Zarembka

, ed. Frontiers in Econometrics. New York: Academic Press; 1974.

37.

Urban

. Urban’s solution (minimum normit x²). In: Bock

Jones

, eds. The Measurement and Prediction of Judgment and Choice. San Francisco: Holden-Day; 1968.

38.

Harris

. The Value of Life. New York: Routledge & Kegan Paul; 1985.

39.

Williams

. Intergenerational equity: an exploration of the “fair innings” argument. Health Econ. 1997;6(2):117–32.

40.

Prosser

Hammitt

Keren

. Measuring health preferences for use in cost-utility and cost-benefit analyses of interventions in children: theoretical and methodological considerations. Pharmacoeconomics. 2007;25(9):713–26.

41.

Craig

Reeve

Brown

. US valuation of health outcomes measured using the PROMIS-29. Value Health. 2014;17(8):846–53.

42.

Craig

Pickard

Stolk

Brazier

. US valuation of the SF-6D. Med Decis Making. 2013:33(6):793–803.

43.

Ramos-Goni

Errea

Rivero-Arias

Cabases

Pinto

. EQ-5D-5L valuation project for the Spanish population: a descriptive overview and preliminary results. Value Health. 2012;15(7):A279–A.

44.

Eidt-Koch

Mittendorf

Greiner

. Cross-sectional validity of the EQ-5D-Y as a generic health outcome instrument in children and adolescents with cystic fibrosis in Germany. BMC Pediatr. 2009;9:55.

45.

Jelsma

. A comparison of the performance of the EQ-5D and the EQ-5D-Y health-related quality of life instruments in South African children. Int J Rehabil Res. 2010;33(2):172–7.

46.

Craig

Runge

Rand-Hendriksen

Ramos-Goñi

Oppe

. Learning and satisficing: an analysis of sequence effects in health valuation. Value Health. In press.

47.

Arnesen

Trommald

. Are QALYs based on time trade-off comparable? A systematic review of TTO methodologies. Health Econ. 2005;14(1):39–53.

48.

Neumann

Weinstein

. Legislating against use of cost-effectiveness information. New Engl J Med. 2010;363(16):1495–7.

49.

Gold

Stevenson

Fryback

. HALYs and QALYs and DALYs, oh my: similarities and differences in summary measures of population health. Ann Rev Pub Health. 2002;23:115–34.

50.

Rushby

Hanson

. Calculating and presenting disability adjusted life years (DALYs) in cost-effectiveness analysis. Health Policy Plan. 2001;16(3):326–31.

51.

Sassi

. Calculating QALYs, comparing QALY and DALY calculations. Health Policy Plan. 2006;21(5):402–8.

52.

Ungar

Prosser

Burnett

. Values and evidence colliding: health technology assessment in child health. Expert Rev Pharmacoecon Outcomes Res. 2013;13(4):417–9.

53.

Basch

Aronson

Berg

. Methodological standards and patient-centeredness in comparative effectiveness research: the PCORI perspective. JAMA. 2012;307(15):1636–40.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.27 MB

0.35 MB

0.26 MB

0.29 MB

0.37 MB

0.23 MB

0.01 MB

0.02 MB

0.01 MB