Psychosocial Risk Profiles Among American and Dutch Families Affected by Pediatric Cancer

Abstract

Objective

Little is known about relations between domains of psychosocial risk among pediatric cancer populations. The Psychosocial Assessment Tool 2.0 (PAT2.0) is one internationally validated screening measure that can examine these relations. This study aimed to examine risk profiles and predictors of these patterns exhibited by American and Dutch families.

Methods

Caregivers of children newly diagnosed with cancer (N = 262; n_USA=145, n_NL=117) completed the PAT2.0 as part of larger studies conducted in the United States and the Netherlands. Latent profile analysis and multinomial logistic regression examined differences in demographic and medical variables across risk profiles. Domains assessed included Family Structure/Resources, Child Problems, Sibling Problems, Family Problems, Caregiver Stress Reactions, and Family Beliefs.

Results

Four groups were identified: “Low-Risk” (n = 162) defined by generally low risk across domains; “Moderate-Caregiver” (n = 55) defined by elevated Caregiver Stress Reactions domain; “Moderate-Children” (n = 25) defined by elevated Child Problems and/or Sibling Problems, and “Elevated-Risk” (n = 20) marked by generally high overall risk. Dutch families had higher odds of being in the Elevated-Risk group, compared to the Low-Risk group. Caregiver age, gender, and educational attainment predicted group membership. Families classified as Targeted or Clinical had higher odds of being in the Moderate or Elevated risk groups.

Conclusion

The PAT2.0 appears to identify largely similar patterns of risk, suggesting that families experience common psychosocial difficulties in both American and Dutch societies. The two Moderate groups demonstrated specific risk sources, suggesting that evaluation of domain patterns, rather than reliance on PAT2.0 risk level, could be of clinical benefit.

Introduction

Pediatric cancer diagnoses are associated with many stressful experiences and uncertain outcomes, including painful procedures, lengthy hospitalizations, treatment-related side effects, and disruptions to normative roles (Rodriguez et al., 2012). These experiences are potentially traumatic, leading to systemic difficulties among many families (Kazak et al., 2003). Most children exhibit resiliency following their cancer diagnosis, yet a subset evidence significant psychosocial adjustment issues (Graf, Bergstraesser, & Landolt, 2013; Kwak et al., 2013; McCarthy et al., 2016a). Posttraumatic stress also appears to be relatively prevalent among caregivers and siblings, suggesting that a pediatric cancer diagnosis and its treatment has extensive family consequences (Dunn et al., 2012; Kazak, Boeving, Alderfer, Hwang, & Reilly, 2005; Long et al., 2018; Price, Kassam-Adams, Alderfer, Christofferson, & Kazak, 2016). This distress is observable early in the treatment process, and identification of those at risk is crucial for prevention and comprehensive care (Dunn et al., 2012; Kazak et al., 2005).

As such, screening for psychosocial risk among families affected by pediatric cancer is essential. Psychosocial screening is a broad public health method for empirically ascertaining those who are most likely to need follow-up care (Kazak, Schneider, Didonato, & Pai, 2015; Kazak et al., 2012). It is a systematic, brief, and cost-effective approach to preventing negative outcomes and limiting exacerbations (Kazak et al., 2012). The Pediatric Psychosocial Preventative Health Model (PPPHM) provides a three-tiered framework for understanding how screening can distinguish the degree of risk for maladjustment related to a pediatric medical condition (Kazak, 2006; Kazak et al., 2015). This model suggests that screening is vital for identifying those with “clinical” levels of risk as well as for monitoring those in the “targeted” range in order to prevent later difficulties. Thus, screening is the first step toward more thorough assessment and early intervention (Kazak, 2006). In fact, the recently established Psychosocial Standards of Care for Children with Cancer and Their Families included assessment as a necessary component of comprehensive care, and underscored the importance of early, frequent, and routine screening (Kazak et al., 2015). Research shows that screening is an acceptable and feasible process that both parents and healthcare providers recognize as beneficial and important for improved communication (Di Battista et al., 2015; McCarthy et al., 2016a, 2016b; Pierce et al., 2017; Sint Nicolaas et al., 2016).

The Psychosocial Assessment Tool 2.0 (PAT2.0) is one measure that facilitates family-centered psychosocial screening in pediatric oncology clinics (Pai et al., 2008). The PAT2.0 uniquely assesses several cross-cutting issues, including problems among multiple family members, and difficulties in social domains affecting functioning and healthcare (Kazak et al., 2001; Pai et al., 2008). Based on the PPPHM, the PAT2.0 categorizes most families (e.g., 55% by maternal report) as “Universal”, suggesting resiliency and adaptability (Pai et al., 2008). The “Targeted” (e.g., 32% by maternal) and “Clinical” (e.g., 13% by maternal report) groups are smaller subsets, consisting of families with increasing risk for severe or escalating difficulties (Pai et al., 2008). Importantly, the PAT2.0 has also been extensively validated with adaptations and translations in several countries including the Netherlands (NL), Canada, and Australia, demonstrating the far-reaching utility of this screening measure (Barrera et al., 2014; McCarthy et al., 2016b; Pai et al., 2008; Sint Nicolaas et al., 2016). The guidelines set by Beaton et al. (2000) have been upheld in the adaptation process, resulting in international versions of the PAT2.0 that were developed to reflect both linguistic and cultural equivalency.

Indeed, it is important to consider social and cultural differences, in addition to linguistic differences, when assessing familial psychosocial risk across cultures. For instance, the Dutch healthcare system differs significantly from the American healthcare system in which the PAT2.0 was initially validated (Sint Nicolaas et al., 2016). Specifically, Dutch citizens are required to have health insurance, are given paid sick leave, have greater outpatient medical management, and often have shorter commutes for care (Euro Health Consumer Index (EHCI), 2014; Moss, 2013). Such factors are known to relate to healthcare utilization, illness beliefs and management behaviors, and even health outcomes, thus underscoring the need to evaluate the possible impact of culture and social environment when utilizing assessment measures across cultures (McQuaid, 2008; Van Widenfelt, Treffers, De Beurs, Siebelink, & Koudijs, 2005). Therefore, examining potential differences and similarities in response patterns to such screening could bolster our understanding of factors that may be universally associated with adjustment outcomes and those that may be culture-dependent.

Central to the primary aim of this article, very little is currently known about patterns of risk experienced by families across domains. A more robust understanding of how risk in one domain relates to risk in other domains may suggest targets for intervention or monitoring among families with varying levels of overall risk. Moreover, the identification of relationships in previous research, such as between sibling distress and family functioning, or between parent and patient distress, would suggest that an examination of these relationships among PAT2.0 risk domains is needed (Bakula et al., 2019; Van Schoors et al., 2017). Data are also lacking regarding potential differences in family risk patterns across cultures, as only the overall distribution across risk tiers has been directly compared between Dutch and American samples (Sint Nicolaas et al., 2016). Overall risk appears similar, with 68% of Dutch mothers, as compared to 55% of American mothers, and 63% of Dutch fathers, as compared to 67% of American fathers reporting their families to be in the Universal tier (Sint Nicolaas et al., 2016). Consequently, exploring associations across domains of psychosocial risk, rather than relying on the three-tiered that classifies families’ overall risk as is often done in practice, appears important (Kazak et al., 2001; Pierce et al., 2017).

Thus, this study aimed to identify latent patterns of psychosocial risk experienced by American and Dutch families affected by pediatric cancer. The present goal was to ascertain relationships between risk domains, based on domains measured by the PAT2.0, and to examine if family nationality was linked with specific vulnerabilities. It was anticipated that several latent risk profiles would be observed, distinguished by related elevations in specific domains. Additionally, it was hypothesized that nationality would not predict risk profile, as previous validation research suggests that the PAT2.0 identifies psychosocial risk similarly across several nations, with families experiencing comparable levels of overall psychosocial risk. Given previous research suggesting that demographic and disease variables are associated with psychosocial stress, these factors were tested as predictors of risk profiles as an exploratory aim (Pinquart, 2017). PAT2.0 risk classification (i.e., Universal, Targeted, Clinical) was also examined as a predictor of latent class membership, with the hypothesis that PAT 2.0 risk classification would be a strong predictor of risk profile membership.

Methods

Participants and Procedures

Of note, recruitment procedures and sample characteristics differed between the Dutch and American studies utilized in the present analysis. Methodological processes are described for each study independently, and statistical differences between samples can be found in Table I.

Table I.

Demographic and Illness Information of Sample (N = 262)

	United States (n = 145)		NL (n = 117)		t-test
Variables	N/M	%/SD	N/M	%/SD	p-value
Patient characteristics
Gender					.806
Female	61	42.1%	51	43.6%
Race/ethnicity					<.001
Caucasian/Dutch	123	84.8%	117	100%
African-American	13	9.0%
Asian	3	2.1%
Native American	8	5.5%
Hispanic/Latino	7	4.8%
Multiracial	4	2.8%
Other	2	1.4%
Age (years)	7.01	5.18	8.48	5.15	.033
Diagnosis
Hematological
Leukemia	55	37.9%	29	24.8%
Hodgkin’s lymphoma	7	4.8%	8	6.8%
Non-Hodgkin lymphoma	7	4.8%	9	7.7%
Other leukemia/lymphoma	6	4.2%	5	4.3%
Neuro-oncological
Brain/CNS tumor	20	13.8%	16	13.7%
Solid tumor
Neuroblastoma	5	3.4%	8	6.8%
Ewing’s sarcoma	2	1.4%	6	5.1%
Rhabdomyosarcoma	8	5.5%	7	6.0%
Wilm’s tumor	14	9.7%	6	5.1%
Osteosarcoma	5	3.4%	11	9.4%
Other solid tumor	14	9.7%	11	9.4%
Time since diagnosis (days)	57.4	35.06	41.7	68.8	<.001
Caregiver characteristics
Gender					<.001
Female	126	86.9%	77	65.8%
Age (years)	36.0	8.5	40.4	6.5	<.001
Marital status					<.001
Married/partnered	111	76.6%	114	97.4%
Separated/divorced	13	9.0%	2	1.71%
Single	18	12.4%	1	0.9%
Other	2	1.4%
Education					.013
Low	35	24.1%	9	7.7%
Medium	31	21.4%	38	32.5%
High	78	53.8%	70	59.8%

Note. Education level of the parent was classified according to the International Standard Classification of Education: low = level 0–2, medium = level 3–5, and high = level 6–8.

Netherlands

Families of children within one year of a cancer diagnosis (N = 117) were consented in clinics at three pediatric oncology centers in the NL. Parents were included if they had a child diagnosed with a first occurrence of cancer prior to age 19 (M_age = 8.38, SD = 5.15; M_{days since diagnosis} = 41.7, SD = 68.80), were receiving curative treatment, and spoke fluent Dutch. One parent for each child registered for an online questionnaire website, where they completed an electronic version of the PAT2.0, among a larger battery of measures. The NL sample had a 59% response rate. All medical ethics committees of the participating sites approved the study; further methodological details can be found in Sint Nicolaas et al. (2016).

United States (USA)

Families of children within 1 year of a cancer diagnosis (N = 146) were consented in oncology clinics at three large mid-western children’s hospitals. Parents were included if they had child with a confirmed diagnosis of a pediatric cancer prior to age 18 years, were receiving curative treatment, spoke fluent English, and were recruited between 2 and 12 weeks following diagnosis (M_age = 7.01, SD = 5.18; M_{days since diagnosis} = 57.4, SD = 35.06). One caregiver for each child completed the PAT2.0 among a larger battery of questionnaires. At the time of analysis, the U.S. sample had a 66% recruitment rate. Institutional Review Board approval was obtained, and all procedures complied with American Psychological Association ethical standards. Families in the U.S. sample were recruited as part of a randomized control trial. However, only baseline data collected during that early diagnosis period is included in this study.

Measures

Psychosocial Assessment Tool 2.0

The PAT2.0 is an observer-report questionnaire aimed at assessing familial psychosocial risk. It has been shown to have good convergent validity with a variety of relevant measures of distress and has been validated with strong internal consistency across domains (Pai et al., 2008; Sint Nicolaas et al., 2016). The measure is composed of seven subscales including Family Structure and Resources, Family Social Support, Family Problems, Parent Stress Reactions, Family Beliefs, Child Problems (e.g., psychosocial risk experienced by the child with cancer), and Sibling Problems. Each subscale includes several items, which are measured dichotomously as risk or no risk. The seven subscales are then calculated by dividing the number of risk items by the total number of items in the respective domain, yielding a subscale score ranging from 0.0 to 1.0. The subscale scores were then summed to create a total score ranging from 0.00 to 7.00. PAT2.0 total scores between 0.00 and 0.99 are considered to be “universal risk”, those between 1.00 and 1.95 are considered to be “targeted risk” and those 2.0 and above are considered to be “clinical risk”.

For the Child Problems and Sibling Problems domains, scores are initially calculated separately for those children with cancer and their siblings below age two and those two and above. In this study, scores for each separate age group were collapsed into single risk domains of Child Problems and Sibling Problems. The Sibling Problems domain is left blank and not included in calculations if the family does not have other children, or scores were averaged if there were siblings in both age groups.

Minor differences exist between the American PAT2.0 and the Dutch translation of the PAT2.0. These differences include the removal of a question related to health insurance on the Dutch version, due to differences in healthcare systems between the United States and the NL (Government of the Netherlands, 2015). More on the differences between the PAT2.0 forms can be found in Sint Nicolaas et al. (2016). Thus, American and Dutch domain scores were converted to z-scores for all domains in the current analyses. These z-scores were then combined into seven variables encompassing the seven risk domains for the entire sample. Further, it is important to note that a PAT3.0 version is now available, but was not available or translated at the time of the study (Kazak et al., 2018).

Overview of Analyses

First, descriptive analyses were conducted and sample demographics and mean scores were compared using SPSS version 25. Additionally, descriptive statistics detailed variable properties that informed inclusion and constraint decisions for subsequent analyses. Specifically, these descriptive analyses identified unacceptable skewness (>2) in the Social Support domain across the entire sample, resulting in this domain being dropped from subsequent analyses. Next, a latent profile analysis was conducted with z-scores of the remaining six domains, utilizing Mplus version 8.1, to obtain subgroups based on the observed response patterns across the six PAT2.0 domains of the entire sample, including all Dutch and American participants. Using maximum likelihood estimation with robust standard errors, the probability of an individual’s membership in each of the a priori unknown latent classes, and the most likely class membership, is produced. The identified profiles are distinguished by class-specific subscale score patterns, with individual differences in domain scores accounted for by class membership.

Six sequential latent profile models were identified ranging from models with one to six latent classes. Relative model fit was evaluated with the Vuong-Lo-Mendell-Rubin Adjusted Likelihood Ratio Test (VLMR), and the Bootstrapped Parametric Likelihood Ratio Test (BLRT; Nylund, Asparouhov, & Muthén, 2007). P-values <.05 were assumed to indicate that the larger class solution was a better model fit, and preference was given to the BLRT when differences between tests existed (Henson, Reise, & Kim, 2007). Accuracy of classification and model fit was examined with the Akaike Information Criterion (AIC), the Bayesian Information Criterion (BIC), the sample size-adjusted BIC, and entropy. Lower AIC and BIC values and entropy values closer to 1 indicate better fit (Geiser, 2013). Variances were freely estimated across classes. For each model, replication of the best log-likelihood was verified to avoid local maxima. For models with more than two classes, the null model log-likelihood for the BLRT was confirmed to equal the best log-likelihood value of the model with one less class.

Following identification of the optimal latent profile solution, the R3STEP procedure was conducted using Mplus version 8.1 to test for demographic variables related to class membership (Asparouhov & Muthén, 2014). The R3STEP procedure employed classification-error corrected multinomial logistic regression to test if nationality, days since diagnosis, patient and caregiver age and sex, caregiver educational level, and caregiver marital status predicted profile membership. The R3STEP procedure accounts for the probabilistic nature of profile membership, and provides estimates that are the log odds of class membership, based on level of the predictor. Additionally, the PAT2.0 risk level, as measured by the standardized three-tiered system, was also tested as a predictor of risk profile membership in the R3STEP procedure.

Results

The present sample consisted of caregivers of children diagnosed with pediatric cancer from the United States (n = 145) and the NL (n = 117), who were primarily female, Caucasian, and above the age of 35 years, with caregivers in the Dutch sample having an older average age (M_age = 40.4, SD = 6.5; t(260) = −4.65, p < .001) than the American sample (M_age = 36.0 SD = 8.5; see Table I). Overall, leukemia was the most common diagnosis, and children had an average age of 7.62 years, with the Dutch sample having a slightly older average age (t(260) = −2.14, p = .033). Detailed demographics are presented in Table I. The U.S. sample had a mean PAT2.0 total score of 1.24 (SD = 0.78; Targeted), and the Dutch sample had a mean score of 0.80 (SD = 0.62; Universal). Across samples, domain scores ranged from 0.05 to 0.38, with the highest level of risk in the Sibling Problems (below the age of two) domain for the U.S. sample, and in the Child Problems (below the age of two) domain for the Dutch sample. However, it is important to note that only two and nine families, respectively, were included in these domain averages, as not all families consist of children below the age of two. All domain and total PAT z-scores did not differ between the Dutch and American samples (ps > .05). Descriptive statistics for the PAT2.0 domain scores are presented in Table II.

Table II.

Descriptive Statistics of the PAT2.0

		United States (n = 145)			NL (n = 117)
PAT2.0 scale	Scale range	M	SD	Range	M	SD	Range
PAT2.0 total	0.00–7.00	1.24	0.78	0.00–4.07	0.80	0.62	0.00–3.10
1. Family structure/resources		0.19	0.18	0.00–0.71	0.08	0.17	0.00–0.80
2. Social support		0.06	0.15	0.00–1.00	0.05	0.11	0.00–0.50
3. Child problems
<2 years of age (nUSA = 21; nNL = 15)		0.19	0.25	0.00–0.75	0.28	0.23	0.00–0.75
≥2 years of age (nUSA = 127; nNL = 102)		0.25	0.19	0.00–0.89	0.22	0.21	0.00–0.80
4. Sibling problems
<2 years of age (nUSA = 2; nNL = 9)		0.38	0.36	0.13–0.63	0.07	0.09	0.00–0.25
≥2 years of age (nUSA = 92; nNL = 84)		0.15	0.20	0.00–0.78	0.10	0.13	0.00–0.53
<2 and ≥2 of age (nUSA = 13; nNL = 2)		0.17	0.21	0.00–0.55	0.18	0.11	0.11–0.26
5. Family problems		0.24	0.17	0.00–0.80	0.09	0.13	0.00–0.50
6. Stress reaction		0.22	0.27	0.00–1.00	0.12	0.23	0.00–1.00
7. Family beliefs		0.16	0.13	0.00–0.67	0.15	0.18	0.00–0.75

Note. All domain scores are presented as scaled scores based on the version of the PAT2.0 used with each sample, respectively and are not directly comparable. For comparisons, Domain and total z-scores were used and did not differ between the two samples (all ps > .05).

Latent Profile Analysis

According to the VLMR and LMR, the four-class solution had optimal fit to the data and meaningful classes. AIC and BIC values suggested solutions with more classes, but the class sample sizes were too small (<5%) and the LMR values were nonsignificant. See Table III for fit indices. The largest latent profile, labeled the “Low-Risk” group (n = 162, 61.83%), was characterized by low levels of risk, on average, across all domains of the PAT2.0 (Bs < −0.10). A smaller latent profile, termed the “Elevated-Risk” group (n = 20, 7.63%), was characterized by elevated risk, on average, across all psychosocial domains (Bs > 0.10), with particularly high risk, on average, in the Caregiver Stress Reactions domains (B = 2.61, 95% CI [2.38, 2.84]).

Table III.

Fit Statistics for PAT2.0 Domains Latent Profile Analysis

	AIC	BIC	Sample size-adjusted BIC	VLMR	Bootstrap likelihood ratio difference test	Entropy	LMR
1 Class	4294.590	4337.410	4299.365
2 Classes	4123.423	4191.221	4130.983	185.167**	185.167***	.901	180.535**
3 Classes	4039.315	4132.092	4049.660	98.108	98.108***	.855	95.65
4 Classes	3921.532	4039.287	3934.663	87.327*	87.327***	.946	85.142*
5 Classes	3876.246	4018.980	3892.162	59.286	59.286***	.953	57.803
6 Classes	3840.295	4008.007	3858.996	49.95	49.951***	.942	48.702

Note. AIC = Akaike Information Criterion; BIC = Bayesian Information Criterion; LMR = Lo-Mendell-Rubin Adjusted Likelihood Ratio Test; VLMR = Vuong-Lo-Mendell-Rubin Adjusted Likelihood Ratio Test. Bold row was chosen model.

^* p < .05,

^** p < .01,

^*** p < .001.

The optimal model also identified two groups with moderate risk. The third profile labeled “Moderate-Caregiver” group (n = 55, 20.99%), was characterized by moderate risk across domains, which was linked with an elevated Caregiver Stress Reactions domain, on average (B = 0.91, 95% CI [0.84, 0.98]). The fourth profile labeled “Moderate-Children” group (n = 25, 9.52%), was characterized by overall moderate risk, which was linked with significantly elevated risk in the Child Problems (B = 0.90, 95% CI [0.41, 1.39]), and Sibling Problems domains, on average (B = 1.73, 95% CI [1.24, 2.21]). Of note, five families in this group did not contain siblings and were, therefore, classified in this group according to elevated Child Problems scores only. See Figure 1 for a graphical representation of group risk across domains.

Figure 1.

Mean domain z-scores by PAT2.0 latent profiles.

Note. Not all families contain siblings, thus the Sibling Problems domain is a representation of the mean score for those families in each group that do contain Siblings.

Predictors of Class Membership

Predictors of class membership were examined via the R3STEP procedure. Patient demographic and disease variables did not significantly predict group membership.

Nationality

Nationality predicted membership in the Elevated-Risk group, with Dutch nationality associated with higher odds of being in the Elevated-Risk group (B = 2.39, SE = 1.19, p = .04) compared to the Low-Risk group. Nationality did not predict membership in any other groups, relative to the Low-Risk group, Elevated-Risk group, or Moderate-Children group.

Caregiver Demographics

Caregiver sex predicted membership in the Moderate-Caregiver group, with male sex associated with lower odds of being in the Moderate-Caregiver group relative to the Low-Risk group (B = −2.67, SE = 0.78, p = .001). Female sex was associated with higher odds (B = −2.79, SE = 1.10, p = .01) of being in the Moderate-Children group than in the Low-Risk group. Greater caregiver age was associated with higher odds (B = 0.14, SE = 0.05, p = .009) of being in the Moderate-Children group than in the Low-Risk group. Higher caregiver educational attainment was associated with greater odds (B = 1.05, SE = 0.48, p = .029) of being in the Moderate-Caregiver group rather than the Elevated-Risk group. Further, greater caregiver age (B = 0.11, SE = 0.04, p = .012) and lower educational attainment (B = −0.99, SE = 0.39, p = .010) were associated with higher odds of being in the Moderate-Children group than the Moderate-Caregiver group. No other demographic or disease characteristics predicted group membership relative to the Elevated-Risk group or Moderate-Children group.

PAT2.0 Risk Classification

Original PAT2.0 risk classification (i.e., Universal, Targeted, Clinical), based on all seven PAT2.0 domains, predicted group membership, with higher risk categories (e.g., Targeted and Clinical) associated with greater odds of being in the Moderate-Children (B = 5.82, SE = 1.02, p < .001), Moderate-Caregiver (B = 4.39, SE = 0.89, p < .001), and the Elevated-Risk groups (B = 7.35, SE = 1.16, p < .001), relative to the Low-Risk group (see Figure 2). Additionally, higher PAT2.0 risk classification was associated with higher risk of being in the Elevated-Risk relative to the Moderate-Caregiver group (B = 2.96, SE = 0.77, p < .001) and the Moderate-Children group (B = 1.53, SE = 0.70, p = .028). Lastly, families with higher PAT2.0 risk classifications were more likely to be in the Moderate-Children group than in the Moderate-Caregiver group (B = 1.43, SE = 0.61, p = .018). This suggests that PAT2.0 risk classification is aligned with the latent profiles, with those in the Universal classification most likely to be in the Low-Risk group and those in the Targeted and Clinical classifications most likely to be in the Moderate and Elevated-Risk groups. However, PAT2.0 risk classification was not an exact predictor (e.g., all of those in the Low-Risk group were not classified as Universal), with 17.3% of those in the Low-Risk group classified as Targeted, 45% of those in the Elevated-Risk group classified as Targeted, and 16.4% of the Moderate-Caregiver group and 32% of the Moderate-Children group classified as Clinical.

Figure 2.

PAT2.0 risk classification by latent profile.

Discussion

To our knowledge, this is the first study to directly examine the pattern of psychosocial risk experienced across domains by families affected by pediatric cancer across two countries. We also sought to identify demographic and disease variables associated with membership in different groups. Latent profile analysis identified four groups that differed in their overall risk across PAT2.0 domains, with particularly salient differences in specific and related sources of psychosocial risk. Low-Risk and Elevated-Risk groups were defined by generally low levels of risk across PAT2.0 domains and generally high levels of risk, respectively. Interestingly, two groups characterized by generally moderate levels of risk were identified but were differentiated by specific elevation patterns. Namely, the Moderate-Caregiver group was defined by an elevated Caregiver Stress Reactions domain, whereas the Moderate-Children group was defined by elevations in Child Problems and Sibling Problems when families contained siblings.

The present findings also indicated that families affected by pediatric cancer experience similar cross-culturally relevant patterns of psychosocial risk. Although Dutch nationality was associated with greater odds of being in the Elevated-Risk group relative to the Low-Risk group, nationality was not related to the likelihood of being a member of any other group. Due to differences in parental education attainment across the two samples, it may be that the larger proportion of highly educated Dutch parents may render this sample vulnerable to greater psychosocial risk. However, this interpretation should be considered in light of the finding that greater parental educational attainment was associated with likely membership in the Moderate-Caregiver group, and not in the Elevated group. Overall, despite differences in the healthcare systems and sociocultural environments, families in both countries seemed to experience common risk patterns, which underscores the strong international clinical utility of the PAT2.0 (EHCI, 2014; Moss, 2013). It appears that factors commonly known to influence the cancer experience outweigh differences in national sociocultural environments as determinants of family psychosocial risk, a finding that may benefit the growing understanding of how to support families worldwide. This finding also highlights the continued need for international collaborations, with the aim of developing family-centered assessment methods and improving comprehensive care with tools that can be appropriately adapted and employed on a larger global scale (Van Widenfelt et al., 2005). Indeed, the value of not only translating measures but also adapting them for use in unique cultural contexts is essential regardless of similarities in symptom presentation (Hambleton, 2001).

Although PAT2.0 risk classification did predict group membership, our study also suggests that it is important to evaluate the individual domains of the PAT2.0 or patterns of domain scores to ascertain a comprehensive picture of risk status, rather than solely relying on the risk level that stems from the Total Score. PAT2.0 risk classification has been shown to predict later distress, yet the distinct moderate groups, as well as the heterogeneity of risk categorization within groups, indicates that the three-tiered PPPHM risk model may not illustrate the full extent of difficulties experienced in particular domains (Kazak et al., 2001; Schepers et al., 2018). Of note, the latent patterns identified in this study did not include evaluation of the Social Support domain, whereas the PAT2.0 risk classification accounts for this area of risk. Although this domain appears to have a limited effect on Total Score, it may have contributed to discrepancies found between PAT2.0 risk classification and group membership.

However, less is known about relationships between specific domains and later relevant outcomes, which may be important for families in the moderate groups who have domain-specific vulnerabilities that could be overlooked due to the lack of global risk. For instance, those families with elevated Child Problems were likely at risk for elevated Sibling Problems (e.g., the “Moderate-Children” group), despite moderate levels of family risk. This indicates that it may still be important to provide family-based psychosocial care to address the needs of both patients and siblings in this group. Interventions that incorporate all family members and address the unique risk that each member may be facing have been shown to be effective and may be indicated in this context (Kazak et al., 2004). Alternatively, it may be that families in the moderate groups could experience escalation of risk, due to risk in one domain. For instance, families in the Moderate-Caregiver group may be at increased risk for the development of adjustment difficulties among children with cancer, due to the well-established link between parent and child distress (Bakula et al., 2019). Early intervention for parents of children with cancer has demonstrated positive results for both parents and children, suggesting that higher levels of care may be valuable for preventing escalation in this moderate group (Fedele et al., 2013). Thus, our findings suggest that it is clinically useful to evaluate a family’s pattern across domains, or the relationship between risk in various domains, in order to more readily target specific sources of psychosocial risk.

Importantly, several demographic characteristics were identified as predictors of group membership. Female caregivers had greater odds of being in the moderate groups, compared to the Low-Risk group. Older caregivers had greater odds of being in the Moderate-Children group, relative to the Low-Risk group or the Moderate-Caregiver group. Consistent with previous literature demonstrating that mothers report greater distress than fathers, this study found that female caregivers are more vulnerable to moderate elevations, with older caregivers reporting significant risk among their children (Pai et al., 2007). The potential role of differences in caregiver age and gender between the Dutch and American samples should also be considered, as these factors may hold differing meaning for psychosocial risk in the two cultures. Higher caregiver educational attainment was also associated with greater odds of being in the Moderate-Caregiver group, relative to the Elevated-Risk group and the Moderate-Children group. This suggests that caregivers with greater education are at risk for higher stress reactions following a pediatric cancer diagnosis, but lower broad familial psychosocial risk. Due to differences in education levels across samples, it is important to note that the Dutch education system differs substantially from the American system, with broader accessibility and federally subsidized higher education (Johnstone & Marcucci, 2010). Thus, for Dutch caregivers education may be less associated with socioeconomic status, employment, and other factors that buffer against parental distress. Future studies are needed to explore the effect of these demographic characteristics on familial risk patterns across cultures, as well as the potentially interactive effect between various predictors.

The domain that most notably distinguished groups was Caregiver Stress Reactions, which evaluates medical traumatic stress symptomology, with the Elevated-Risk and Moderate-Caregiver groups showing elevations in this domain. As it is well-documented that caregivers are at risk for medical traumatic stress, this domain may be particularly salient as an identifier of risk profiles (Dunn et al., 2012; Kazak, Boeving, Alderfer, Hwang, & Reilly, 2005). The significant difference in days since diagnosis between the Dutch and American samples may also influence the present findings, suggesting the need for closer examination of the emergence of medical traumatic stress symptomology. More broadly, this finding aligns with previous research suggesting that cognitive appraisals have an important role in adjustment to a pediatric diagnosis, perhaps even more so than disease or demographic factors, which demonstrated limited relationships with group membership (Mullins, Cote, Fuemmeler, Jean, Beatty, & Paul, 2001; Szulczewski, Mullins, Bidwell, Eddington, & Pai, 2017). Interestingly, the Family Beliefs and the Family Structure and Resources domains did not appear to play a significant role in differentiating groups. Further examination of how these specific domains relate to the other PAT2.0 domains, and how they might predict future distress, is warranted.

Although this study addresses an important gap in the literature, there are several limitations. First, there were significant demographic differences between the American and Dutch samples (e.g., the Dutch sample had older caregivers and children, on average) and different methodologies were used to recruit participants and collect the PAT2.0 data. Parent and child age are known factors associated with psychosocial outcomes, and differences in age may have played a role in the present results. The potential contribution of recruitment procedures and rates to the current findings must also be considered, as a variety of familial factors may influence the decision to participate in different study designs. Both samples were also predominantly Caucasian, necessitating future research among more diverse samples. Second, demographic and disease characteristics were parent-reported, which may result in misclassification of disease and time since diagnosis. Further, all risk data were parent-reported, which may result in reporting bias. It is also important to note that the Social Support domain had to be dropped from analyses due to unacceptable skewness related to very low levels of risk in this domain across the samples. Thus, dropping this domain prevented evaluation of how social support difficulties may relate to risk in other domains. Further, this study did not include any measures for comparison or follow-up distress data among the American sample and, therefore, the relationship between the presently identified risk profiles and later psychosocial distress cannot be determined. Lastly, the updated PAT3.0 has not yet been adapted for use in the NL, precluding analysis of family patterns based on this measure. The revisions were particularly pertinent for families of younger patients and broader literacy levels, necessitating future examination of the present results with the newer screening tool (Kazak et al., 2018).

Given these limitations, it is clear that replication of these findings is necessary to confirm the optimal model solution. Cross-cultural comparisons with non-Western countries and other translated versions of the PAT2.0 would also provide greater information on the applicability of these risk profiles across other cultures. Overall, this study further validates the use of the PAT2.0 across the United States and the NL, while highlighting the clinical and research importance of examining the domain scores of psychosocial risk in an effort to enhance tailored assessment and intervention for families affected by pediatric cancer.