Empirically derived profiles of neurocognitive functioning in youth and young adults with sickle cell disease

Abstract

Objective

Sickle cell disease (SCD) is an inherited blood disorder associated with neurocognitive deficits. In contrast to variable-centered approaches, no known research has utilized person-centered strategies to identify multidimensional patterns of neurocognitive functioning of an individual with SCD. The purpose of the present study was to create empirically derived profiles and identify predictors of neurocognitive functioning subgroups among youth and young adults with SCD.

Methods

Individuals with SCD (N = 393, mean age 14.05 years, age range 8–24, 50.4% female/49.6% male) completed neurocognitive assessments. Latent profile analysis derived subgroups/classes of neurocognitive functioning and determined relations with demographic and medical variables.

Results

Three latent classes emerged: average functioning (n = 102, 27%), low average functioning (n = 225, 60%), and exceptionally low functioning (n = 46, 12%). Older age was associated with membership in the low average and exceptionally low functioning groups (relative to the average group). Being prescribed hydroxyurea was associated with membership in the average functioning group (relative to the low average group) and absence of hydroxyurea use was associated with membership in the exceptionally low group (relative to the low average group). Lower social vulnerability was associated with membership in the average functioning group compared to the low average and exceptionally low groups.

Conclusions

Clinicians can help reduce disparities in cognitive development for individuals with SCD by promoting early treatment with hydroxyurea and implementing methods to reduce social vulnerabilities that can interfere with access to evidence-based care.

Introduction

Sickle cell disease (SCD) is the most common inherited blood disorder (National Heart, Lung, and Blood Institute, 2021) and affects an estimated 100,000 Americans (Centers for Disease Control and Prevention, 2021). Sickle cell disease primarily impacts those of African descent (Hassell, 2010) and is comprised of several genotypes. Typically, individuals with the SS/SB⁰ genotypes experience more severe complications than those with HbSC/HbSβ⁺-thalassemia or other genotypes (Piel et al., 2013). Sickle cell disease leads to the polymerization of sickle hemoglobin, which interferes with the oxygen carrying capacity of red blood cells and can cause vaso-occlusion (Sundd et al., 2019). Patients with SCD are at increased risk of numerous complications such as pain, jaundice, fatigue, dactilitis, and infections. The brain requires a continuous supply of oxygen and blood glucose for normal function. If the supply is reduced, then the brain cells cease to function normally, and patients develop cerebral infarction (Stotesbury et al., 2019). Thus, individuals with SCD are at an elevated risk of experiencing brain complications in the brain, such as silent infarction or stroke which negatively impact neurocognitive performance (Wang et al., 2001).

Neurocognitive impairment is common among individuals with SCD, particularly those with history of a cerebrovascular accident, including a silent infarct or stroke. A meta-analysis of cognitive function among individuals with SCD found a stepwise increase in cognitive deficits among groups with no infarct, silent infarct, or overt stroke (Prussien et al., 2019). These cognitive deficits become apparent in childhood and worsen with age (Wang et al., 2001). Youth with SCD display deficits in verbal and nonverbal reasoning, graphomotor control, working memory, and processing speed (Hijmans et al., 2011; Schatz & McClellan, 2006). There is a greater risk of neurocognitive deficits in children with lower hemoglobin (Hb) oxygen saturation and lower total Hb (Prussien et al., 2020; Stotesbury et al., 2018). Social and other determinants of health, such as parent education and neighborhood resources, play a role in neurocognitive functioning in youth (King, Strouse, et al., 2014; Prussien et al., 2019). Although neurocognitive functioning is connected to disease-specific processes, it is essential to consider social and other factors that impact health. These socio-demographic factors are proxies for differential restrictions and access to resources, as well as facilitators and barriers at the institutional, interpersonal, and neighborhood levels based on a person’s intersectional identities, such as their race, income, gender, and geographic region (David & Derthick, 2017; National Academies of Sciences, Engineering, & Medicine, 2017). It is crucial to consider these broader determinants of health alongside disease-specific processes to gain a more comprehensive understanding of the factors affecting neurocognitive functioning.

Social vulnerability refers to the social practices and structures in which people live that render them unable to respond or adapt to harms (Li et al., 2023). Some examples of factors that might affect a person’s social vulnerability include household composition, minority status, and access to transportation (Cutter et al., 2003). A broad range of socioeconomic risks affect neurocognitive outcomes in SCD across the lifespan, typically to a degree that is greater than medical variables (Ampomah et al., 2022; King, Rodeghier, et al., 2014). Specifically, social vulnerability uniquely contribute to neurocognitive outcomes in SCD (Heitzer et al., 2023) as well as to many other health-related quality of life outcomes (e.g., pain outcomes, hospital admissions) (Khan et al., 2023).

Hydroxyurea, an advancement of medical treatment in SCD has reduced the risk of stroke, mortality, and pain episodes (Lê et al., 2015; Lopes De Castro Lobo et al., 2013). Hydroxyurea increases the amount of fetal hemoglobin in individuals with SCD, helping prevent the formation of sickle-shaped red blood cells (Fitzhugh et al., 2015). Disease-modifying treatments such as hydroxyurea display preliminary evidence that they are positively associated with neurocognitive functioning (Heitzer et al., 2021); therefore, early treatment can potentially limit later neurocognitive deficits.

Given the likely heterogeneity in neurocognitive functioning among individuals living with SCD, person-centered analyses may be helpful to find subgroups with similar neurocognitive profiles; and if found, what may predict an individual’s membership in those subgroups. Latent profile analysis (LPA) is an exploratory person-centered analytic approach that uses information about the population to separate it into more homogenous subgroups, called latent classes (Berlin et al., 2013). Latent profile analysis can explore profiles of functioning/well-being across different domains (Berlin et al., 2013). However, to date, research has generally used variable-centered approaches to understanding neurocognitive functioning in this population (Prussien et al., 2019). These studies typically explore average neurocognitive performance among patients to determine deficits representing individuals with SCD. While latent profiles of health-related quality of life have been explored (Keenan et al., 2020) in youth and young adults with SCD, person-centered analytic approaches have not been used to explore the multidimensional construct of neurocognitive functioning of those with SCD.

Existing literature has predominantly embraced a deficit-oriented perspective, assuming a generalized neurocognitive impairment among individuals with SCD, without delving into the potential subgroups and the factors that may distinguish these patterns. Our study aimed to fill this gap by introducing a person-centered analytic approach, specifically focusing on the multidimensional construct of neurocognitive functioning. By doing so, we contribute to a more comprehensive understanding of the heterogeneity within the SCD population, moving beyond a broad deficit assumption to identify distinct neurocognitive profiles and potential predictors of subgroup membership. This person-centered approach is crucial for tailoring interventions and support strategies to specific subgroups within the SCD population, acknowledging that individualized care may be more effective than a one-size-fits-all approach. Additionally, these analyses facilitate a shift from a deficit-oriented perspective to a more nuanced exploration of the diverse neurocognitive experiences within the SCD community.

In the present study, LPA was used to examine latent profiles to determine whether subgroups of neurocognitive functioning exist within the broader population and to explore how membership in those subgroups was predicted by medical (use of hydroxyurea, SCD genotype, total Hb concentration) and demographic variables (age, socioeconomic status). Neurocognitive measures were chosen based on a mix of broad domains (e.g., IQ, processing speed, working memory) frequently assessed in the literature (Prussien et al., 2019, 2020) as well as more specific skills (e.g., visual motor and verbal memory) with clear functional correlates. By delving into these variables, the study sought to establish meaningful connections between the diverse characteristics of the study participants and their neurocognitive profiles. This approach allowed for a more nuanced understanding of the factors contributing to the heterogeneity observed in the neurocognitive functioning of individuals with SCD. The objectives of the study were to (1) examine empirically derived subgroups of neurocognitive functioning based on profiles of SCD and (2) to determine demographic and health-related variables that predict membership in SCD subgroups.

Methods

Participants and procedure

Data were collected from St Jude Children’s Research Hospital as part of the Sickle Cell Clinical Research Intervention Program (SCCRIP; Hankins et al., 2018). SCCRIP is a longitudinal lifetime cohort study that gathers data on the progression of psychosocial, neurocognitive, and health outcomes in youth and young adults with SCD. Participants between the ages of 8 and 24 completed neurocognitive testing as part of a cognitive surveillance program. All patients with SCD are referred for targeted neurocognitive assessments every 4 years regardless of disease status or clinical concern. These assessments are performed in four stages: school age (age 8–9), early adolescence (age 12–13), late adolescence (age 16–17), and young adulthood age (19–24). Data were obtained from 391 individuals participating in SCCRIP. If patients had multiple neurocognitive assessments, then the most recent assessment was used. Informed consent was obtained from all eligible participants or their legal guardians if participant was a minor. The institutional review board at St Jude Children’s Research Hospital approved this study.

Demographic, medical, and treatment variables

Demographic, medical, and treatment variables, including sex, age, SCD genotype, and hydroxyurea usage, were extracted from patients’ medical records in the SCCRIP database. Genotypes were dichotomized such that patients with HbSS or HbSB⁰ thalassemia were coded as having a severe genotype whereas those with HbSC, HbSB⁺, or any other genotype were coded as having a less severe genotype. The Social Vulnerability Index (SVI) (Cutter et al., 2003; Flanagan et al., 2018) was used to classify individuals based on social vulnerabilities at the neighborhood level (e.g., poverty, education, housing data) (Cutter et al., 2003). A higher percentile score indicates higher social vulnerability. Participants with HbSS/HbSβ⁰-thalassemia received hydroxyurea according to NHLBI guidelines (Yawn et al., 2014). For participants with HbSC/HbSβ⁺-thalassemia, hydroxyurea initiation was guided by the frequency of acute disease complications (Luchtman‐Jones et al., 2016). Sickle cell disease genotype was included in the statistical models (described below) to account for differences in hydroxyurea prescribing patterns based on genotype. In this study, participants data were marked as “yes” for hydroxyurea use if they were actively prescribed hydroxyurea at any point within 1 year before their completion of neurocognitive assessments for SCD. Conversely, data were coded as “no” if hydroxyurea was not listed among their active medications at any point during that year. Total Hb concentration values were collected at steady state on the day of neurocognitive testing or were the average value of measurements taken within the 3 months prior to testing.

Neurocognitive measures

Participants completed a battery of neurocognitive measures. A psychologist supervised all assessments. The Wechsler Abbreviated Scale of Intelligence- Second Edition (WASI-II) (Wechsler, 2011) provides a 4-subtest abbreviated Intelligence Quotient (IQ). The test–retest reliability for the 4-subtest IQ is 0.94 for children (ages 6–16 years) and 0.96 for adults (ages 17–90 years). Digit Span (working memory) and Coding (graphomotor processing speed) subtests from the Wechsler Adult Intelligence Scale Fourth Edition (WAIS-IV) (Wechsler, 2014) or Wechsler Intelligence Scale for Children, Fourth Edition (WISC-IV) (Wechsler, 2003) were administered depending on the participant’s age. The Digit Span and Coding subtests from the WAIS-IV have test–retest reliabilities ranging from 0.75 to 0.85, respectively for ages 16–29. On the WISC-IV (ages 6–16 years), the Digit Span and Coding subtests demonstrate test–retest reliabilities ranging from 0.83 to 0.84, respectively. The Story Memory (initial recall) subtest from the Wide Range Assessment of Memory and Learning, Second Edition (WRAML-2) measured verbal memory. The test–retest reliability for the Story Memory subtest is 0.75 (Sheslow & Adams, 2003). Initial and delayed recall on Story Memory demonstrate a strong association (r = 0.91), therefore only initial recall was used. Visual motor integration was examined using the Beery–Buktenica Visual Motor Integration (VMI) Test, Sixth Edition. The test demonstrates a reported inter-rater reliability of 0.92, internal consistency of 0.96, and a test–retest reliability of 0.89 (Beery et al., 2009).

Our group has previously presented neurocognitive outcome data from SCCRIP using variable-centered approaches to understand risk and resilience factors influencing neurocognitive performance in patients with SCD (Heitzer et al., 2021; Heitzer, Longoria, Porter, et al., 2022; Heitzer, Longoria, Rampersaud, et al., 2022; Partanen et al., 2020; Saulsberry‐Abate et al., 2021; Trpchevska et al., 2022). No prior publications have analyzed this neurocognitive data using LPA. Unlike prior analyses, LPA allowed us to examine homogenous subgroups and explore profiles of performance across different domains.

Analytic plan

Latent profile analysis was conducted in Mplus Version 8.3 (Muthén & Muthén, 1998–2017), and maximum likelihood estimation with robust standard errors (MLRs) was used to account for missing data and nonnormal distributions. Prior to analyses, class indicators were scaled to a common metric to reflect population-based z-scores, (e.g., [individual scaled score−10]/3; [individual standard score−100]/15). For demographic and medical variables, approximately 10% of the data were missing and assumed to be missing at random and multiple imputation was conducted following the methods of Graham et al. (2007). Latent profile analysis was used to empirically derive subscales of neurocognitive functioning based on the subgroups of youth and young adults living with SCD. In this model, variations in means and variances were considered across classes, but covariances were not included and were set to be the same across profiles. The best model was tested and selected using consensus methods and model fit-statistics (Berlin et al., 2018).

To determine the optimal solution, the models were compared using the following fit-statistics: Bayesian information criteria (BIC) (Schwarz, 1978), Lo–Mendell–Rubin test (LMR) (Lo et al., 2001), Bootstrap likelihood ratio test (BLRT) (Browne & McNicholas, 2013) and entropy as recommended by (Berlin et al., 2013). The BIC is compared based on size difference; a lower score indicates a better model fit (Kass & Raftery, 1995). The LMR and BLRT provide p-values to determine if there is a statistically significant improvement in fit for the inclusion of one or more classes (e.g., 3 compared to 2). It is preferred to choose the number of classes based on the BLRT when inconsistency is found across these indices (Asparouhov & Muthén, 2014). Entropy provides information about the classification accuracy of all variables in the model; values range from 0 to 1, with the higher indicating greater classification accuracy (Berlin et al., 2013). Entropy is used descriptively rather than for determining the number of classes, as it provides useful information about the chosen model. Univariate entropy provides information about the classification accuracy of individual variables in the model. Model class sizes are reviewed to justify not selecting a model with fewer classes, which would be more parsimonious.

After identifying the most suitable model, we added variables related to patient demographics (age and SVI) and medical history (use of hydroxyurea, SCD genotype, Hb) as predictors of class membership by using R3STEP auxiliary variables and following the techniques of Asparouhov and Muthén (2014). By adding these predictors as auxiliary variables, the composition of the classes remained consistent. This approach aimed to provide a more accurate prediction of the class membership by including additional information related to the factors that distinguish each class.

Results

Table 1 provides an overview of the demographic and clinical characteristics of the sample. Overall, 391 individuals completed neurocognitive testing. Nearly all (99%) of the patients were African American. The sample consisted of 49.60% males and 50.40% females having an average age of 14.05 years (SD = 4.76). The participants SCD genotypes were as follows: 61.32% HbSS/HbSβ⁰, 29.26% HbSC, 7.89% HbSB⁺, and 1.53% other genotypes. More than half (53.70%) of patients were treated with hydroxyurea, starting at an average age of 2.33 years (SD = 0.55). Forty patients had received chronic transfusion therapy, and six patients had a history of an overt stroke. Among those not currently being treated with hydroxyurea, 15 were receiving chronic transfusion therapy.

Table 1.

Univariate correlations of demographic, treatment, and neurocognitive variables.

n = 391	1	2	3	4	5	6	7	8	9	10
1. IQ	87.63 ± 14.20
2. Verbal Memory	0.48	9.60 ± 2.73
3.Visual Motor	0.57	0.20	84.22 ± 14.20
4. Processing Speed	0.42	0.22	0.32	8.08 ± 2.67
5. Working Memory	0.53	0.02	0.39	0.36	8.66 ± 2.72
6. Age	−0.17	−0.14	−0.31	−0.10 ns	−0.15	14.05 ± 4.76
7. Genotype (HbSS/HbSβ0)	−0.07 ns	−0.07 ns	−0.08 ns	−0.13	−0.09 ns	0.02 ns	61.32%
8. Hydroxyurea (on treatment)	0.10	0.05 ns	−0.01 ns	0.02 ns	0.000 ns	0.12 ns	0.45	53.70%
9. Total hemoglobin	0.06 ns	0.05 ns	0.05 ns	0.13	0.07	0.03 ns	−0.68	−0.26	10.07 ± 1.74
10. SVI	−0.21	−0.04 ns	−0.21	−0.14	−0.11	0.13	0.06 ns	−0.01 ns	0.07	0.66 ± 0.25

Note. The intersection of different variables represents their univariate correlation (represented by italicized font). The intersection of the same variables represents their descriptive statistics (interval level data displayed as mean±standard deviation, categorical data displayed as a percentage of the larger sample). NS=p-value ≥.05. IQ=Wechsler Abbreviated Scale of Intelligence, Second Edition, 4-subtest IQ; Verbal Memory=The Wide Range Assessment of Memory and Learning, Second Edition, Story Memory subtest Initial Recall; Visual motor=Beery–Buktenica Visual Motor Integration Test, Sixth Edition; Processing Speed=Wechsler Adult Intelligence Scale, Fourth Edition or Wechsler Intelligence Scales for Children, Fourth/Fifth Edition Coding subtest; Working Memory=Wechsler Adult Intelligence Scale, Fourth Edition or Wechsler Intelligence Scales for Children, Fourth/Fifth Edition Digit Span subtest. IQ and Visual Motor performance displayed as standard scores with a normative mean of 100 and standard deviation of 15. Verbal Memory, Processing Speed, and Working Memory displayed as scaled scores with a normative mean of 10 and standard deviation of 3. The distribution of genotypes included: 61.3% HbSS/HbSβ⁰, 29.3% HbSC, 7.9% HbSB⁺, and 1.5% other genotypes. Social Vulnerability Index (SVI) (Cutter et al., 2003; Flanagan et al., 2018) was used to classify individuals based on social vulnerabilities at the neighborhood level; a higher score indicates greater social vulnerability (scale 0–100).

Table 1 lists univariate correlations between demographic, medical, and neurocognitive variables. Two to six classes were run to determine the optimal fit between classes, see Table 2. The 3-class model was chosen based on BIC, LMR p, and BLRT p. The BIC was the lowest of all models tested. LMR p and BLRT p displayed significant improvement for the 3-class compared to the 2-class and a lack of significant improvement for the 4-class model compared to the 3-class (where BLRT is marginally significant at 0.0000). The loglikelihood of the 3-class model was replicated indicating that global (vs. local) maxima were found. IQ, Visual Motor, and Working Memory contributed the most to the latent class separation with the following univariate respective entropies 0.543, 0.356, and 0.358. Verbal Memory and Processing Speed contributed less information to the latent class variable (0.238 and 0.281 respectively).

Table 2.

Model comparison for latent profile analyses.

Number of classes	Entropy	BIC	LMR p	BLRT p	Smallest class size (n)
2	0.65	4529.31	<.01	<.01	171
3	0.75	4455.34	<.01	<.01	47
4	0.80	4475.62	.16	.07	6
5	0.76	4501.18	.28	.05	6
6	0.77	4536.58	.36	.67	6

Note. The model solution chosen is bolded. BIC=Bayesian Information Criterion compared based on size difference; a lower score indicates a better model fit; LMR=Lo–Mendell–Rubin Test; LMR p=Lo–Mendell–Rubin test p-value determine if there is a statistically significant improvement in fit for the inclusion of one or more classes; BLRT=Bootstrap likelihood ratio test; BLRT p = Bootstrap likelihood ratio test p-value determine if there is a statistically significant improvement in fit for the inclusion of one or more classes. Entropy ranging from 0 to 1 with higher scores representing greater classification accuracy was also explored, though not used to decide between competing class models.

The 3-class model presented distinct classes deemed the average functioning, low average functioning, and exceptionally low functioning groups (Figure 1). The names of the latent classes were based on the American Academy of Clinical Neuropsychology uniform labeling of performance test scores (Guilmette et al., 2020). The average functioning group comprised 27% of the sample (n = 107), with average neurocognitive performance subscale z-scores ranging from IQ (0.09, SE = 0.16), Processing Speed (0.00, SE = 0.10), Working Memory (0.24, SE = 0.13), Verbal Memory (0.37, SE = 0.19), and Visual Motor (−0.41, SE = 0.08). Low average functioning encompassed 60% of the sample (n = 237) with the neurocognitive performance subscale z-score means ranging from IQ (−0.99, SE = 0.07), Processing Speed (−0.78, SE = 0.08), Working Memory (−0.53, SE = 0.07), Verbal Memory (−0.25, SE = 0.08), and Visual Motor (−1.12, SE = 0.10). Exceptionally low functioning encompassed 12% of the sample (n = 47) with the neurocognitive performance subscale z-score means ranging from IQ (−2.09, SE = 0.13), Processing Speed (−1.41, SE = 0.17), Working Memory (−1.53, SE = 0.13), Verbal Memory (−0.98, SE = 0.16), and Visual Motor (−2.62, SE = 0.22). There was a similar pattern of neurocognitive performance across the three classes based on relative performance within classes. The visual motor domain was the lowest domain for all classes, followed by IQ and working memory. The strongest domain was verbal memory for the average and exceptionally low functioning groups.

Figure 1.

Neurocognitive profiles of youth and young adults with sickle cell disease based on latent profile analysis. Dashed-line: exceptionally low functioning class (n = 47, 12%), dotted-line: low average functioning class (n = 237, 60%), average functioning class (n = 107, 27% of study sample), class indicators were scaled to a common metric to reflect population-based z-scores (e.g., [individual scaled score−10]/3; [individual standard score−100]/15). Error bars indicate 95% confidence interval. IQ=Wechsler Abbreviated Scale of Intelligence, Second Edition, 4-subtest IQ; Verbal Memory=Wide Range Assessment of Memory and Learning, Second Edition, Story Memory subtest; Visual Motor=Beery–Buktenica Visual Motor Integration Test, Sixth Edition; Processing Speed=Wechsler Adult Intelligence Scale, Fourth Edition or Wechsler Intelligence Scales for Children, Fourth/Fifth Edition Coding subtest; Working Memory=Wechsler Adult Intelligence Scale, Fourth Edition or Wechsler Intelligence Scales for Children, Fourth/Fifth Edition Digit Span subtest.

Multinomial logistic regression analyses (see Table 3) indicated that age, use of hydroxyurea, and SVI significantly predicted class membership, whereas Hb and SCD genotype did not predict class membership. Older age was associated with an increase in the log odds of being in the low average (0.19, SE = 0.050, p < .01) or exceptionally low (log odds = 0.10, SE = 0.04, p = .02) functioning group rather than the average functioning group. Also, age predicted a small decrease in the log odds (−0.08, SE = 0.04, p = .03) of being in the low average functioning group rather than the exceptionally low functioning group. Hydroxyurea use was associated with an increase in the log odds of being in the average functioning class relative to those of being in the low average class (1.66, SE = 0.54, p < .01), and patients who were not prescribed hydroxyurea had an increase in the log odds of being in the exceptionally low class (1.28, SE = 0.4, p < .01) relative to those of being in the low functioning class. An increase in the SVI z-score was associated with an increase in the log odds of being in the low average (0.79, SE = 0.29, p = .01) or exceptionally low (0.58, SE = 0.21, p = .01) class rather than the average class. The log odds of disease genotype and Hb in the prediction of class membership did not significantly differ from zero.

Table 3.

Comparative demographic and medical variable predictions of latent class membership.

Predicted class	Predictor	Log odds	SE	Log odds/SE	p	Odds ratio=exp (log odds)
Low average functioning class (reference: average functioning class)	Sex	0.64	0.46	1.39	.17	1.89
	Age	0.19	0.05	3.72	<.01*	1.20
	Genotype	1.14	0.80	1.43	.15	3.12
	Hydroxyurea	−1.67	0.54	−3.11	<.01	0.19
	Total hemoglobin	−0.15	0.19	−0.79	.43	0.86
	Social vulnerability	0.79	0.29	2.75	<.01*	2.20
Exceptionally low functioning class (reference: average functioning class)	Sex	0.29	0.34	0.85	.40	1.33
	Age	0.10	0.04	2.37	.02*	1.11
	Genotype	0.45	0.49	0.92	.36	1.57
	Hydroxyurea	−0.38	0.41	−0.93	.35	0.68
	Total hemoglobin	−0.15	0.13	−1.16	.25	0.86
	Social vulnerability	0.58	0.21	2.83	<.01*	1.79
Exceptionally low functioning class (Reference: low average functioning class)	Sex	−0.35	0.41	−0.84	.40	0.71
	Age	−0.08	0.03	−2.24	.03*	0.92
	Genotype	−0.68	0.74	−0.92	.36	0.51
	Hydroxyurea	1.28	0.45	2.87	<.01*	3.61
	Total hemoglobin	0.00	0.18	<0.01	.99	1.00
	Social vulnerability	−0.20	0.26	−0.79	.43	0.82

Note. A one unit increase in the predictor variable results in the estimated log odds/OR of being in the predicted (comparison) class versus the reference class. Redundant comparisons were excluded from the table. Reference class is indicated in parentheses. SE = standard error; log odds=multinominal logistic regression; Social Vulnerability classifies individuals based on social vulnerabilities at the neighborhood level (e.g., poverty, education, housing data); a higher percentile score indicates higher social vulnerability. Sickle cell disease genotype HbSS/HbSβ⁰-thalassemia and HbSC/HbSβ+-thalassemia/Other.

^*Values statistically significant at p < .05.

Discussion

The present study aimed to identify distinct subgroups of neurocognitive functioning in youth and young adults with SCD and to establish whether demographic and medical variables could be used to predict membership in subgroups. The three-class model presented distinct latent classes of average functioning, low average functioning, and exceptionally low functioning. The results indicated similar patterns of relative neurocognitive functioning across latent classes. All groups displayed a relative strength in verbal memory and a weakness in visual motor skills. When exploring the relationship between predictor variables and group membership it was determined that older age was associated with membership in the low average and exceptionally low functioning groups. Being prescribed hydroxyurea was associated with membership in the average functioning group, whereas not being prescribed hydroxyurea was associated with membership in the exceptionally low group (relative to the low average group). Lower SVI was associated with membership in the average functioning group.

In the present study, individuals with SCD exhibited deficits across cognitive domains relative to normative expectations. These cognitive domains covered a wide range of skills frequently assessed in neuropsychological evaluations, allowing for a thorough understanding of overall functioning. Across domains, patients displayed a weakness in visual motor skills and a relative strength in verbal memory. These findings align with prior work by Hijmans et al. (2011), that demonstrated a relative weakness in visual motor skills among individuals with SCD. The current study uncovered a unique finding of a strength in verbal memory among individuals with SCD, which has not been reported in previous research. Across classes, there were clear neurocognitive strengths and weaknesses. This finding suggests the need for interventions targeting areas of weakness such as visual motor functioning for most children with SCD, rather than solely focusing on those with lower cognitive functioning across domains. Overall, the observed results using person centered analyses are consistent with prior research using variable centered analyses.

Previous findings have shown that older age is associated with poor neurocognitive performance and that the use of hydroxyurea is positively correlated with neurocognitive performance in youth and young adults with SCD (Heitzer et al., 2021). These findings are consistent with the present results, which suggest that older age and not being prescribed hydroxyurea were associated with membership in the lower functioning groups. However, the impact of hydroxyurea treatment was not consistent across all classes. Specifically, hydroxyurea treatment differentiated between individuals at the extremes (average and exceptionally low classes), suggesting that it may prevent severe deficits often seen in patients with a stroke and potentially buffered performance for those in the average group if prescribed early in life. Moreover, the current study’s results also support prior findings (Heitzer et al., 2023), demonstrating a negative association between neurocognitive performance and the SVI. Specifically, the present results show that individuals who had a low SVI were more likely to be in the average functioning group. These findings suggest that oppressive systems render families vulnerable, as they face a cumulative impact of reduced support and increased challenges. Children with SCD, beyond facing typical life challenges, also endure additional biopsychosocial burdens due to living with a chronic illness, which can encompass pain, neurocognitive issues, stigma, and racism. These struggles are further compounded by residing in resource-deprived neighborhoods, which predictably worsens the situation, pushing these children into less-advantaged circumstances.

These findings have notable implications for clinical interventions aimed at improving neurocognitive performance in patients with SCD. Implementing a proactive approach to support patient development during the early stages has the potential to reduce neurocognitive deficits that may arise in the future. For instance, addressing visual motor and fine motor difficulties through occupational therapy intervention early in life may limit later difficulties.

Numerous studies have emphasized the benefits of hydroxyurea in treating SCD, confirming its long-term safety (Steinberg et al., 2010). Challenges faced by patients and their families encompass reluctance to take medications, insufficient knowledge about hydroxyurea, skepticism regarding its efficacy, financial constraints, difficult obtaining timely refills, and concerns about potential side effects (Badawy et al., 2017). Additionally, the findings underscore the importance of initiating hydroxyurea treatment at a young age to minimize alterations in neurocognitive performance as individuals age. Healthcare providers should advocate for and assist families in navigating barriers to accessing early treatment.

Finally, the strongest predictor of group membership for the average functioning group was social vulnerability, which measures socioeconomic status at the neighborhood level. This highlights the importance of the broader community environment in cognitive development for patients with SCD. Tailored early intervention programs that address the effect of socioeconomic status on the cognitive development of individuals with SCD may minimize deficits as they age. To effectively address the impact of socioeconomic status, it is crucial to direct attention to neighborhoods, cities, or regions where the most-affected socioeconomic groups are more prevalent. For example, to address the specific needs of individuals with SCD living in areas with high social vulnerability, an intervention could target improving transportation options to alleviate the difficulties faced by families to access essential healthcare services and attending medical appointments.

The current study has several strengths, such as a large sample of patients spanning school age to young adulthood. Additionally, this study stands out as the first to investigate person-centered analytic methods in examining the multi-faceted nature of neurocognitive functioning in individuals with SCD. The study has significant limitations. Neuroimaging data were not available for all participants, preventing the investigation of how silent cerebral infarcts might have influenced group membership. There was no control group to assess whether the neurocognitive performance pattern was unique to the SCD population. The study covers a broad array of cognitive measures across age groups. The large age range allows for examination of age-related effects; however, cognitive domains may look different across age groups, potentially hindering interpretation of the unique cognitive profile within specific age bands. Several commonly measured cognitive abilities are not included in our current framework. Consequently, the findings may not fully encompass the entire spectrum of relevant cognitive skills. Acknowledging this limitation emphasizes the need for caution when generalizing our results to broader cognitive domains. In future studies, incorporating additional measures could create profiles based on a wider range of neurocognitive or academic skills. A larger sample size might enable researchers to identify more specific subgroups with unique neurocognitive profiles.

Results revealed that individuals with SCD displayed similar patterns of neurocognitive functioning across average, low average and exceptionally low functioning groups, with strengths in verbal memory and weaknesses in visual motor skills. The person-centered analyses yielded similar results as prior variable-centered analyses (Prussien et al., 2019). Age, hydroxyurea treatment, and social vulnerability were significant predictors of group membership. Clinicians can play a significant role in reducing disparities in cognitive development for individuals with SCD by promoting early treatment with hydroxyurea and implementing methods to reduce social vulnerabilities that can interfere with access to care. Such steps could promote more equitable outcomes and foster optimal cognitive functioning for individuals with SCD.

Data availability

The data underlying this article will be shared on reasonable request to the corresponding author.

Author contributions

Vinkrya Ellison (Conceptualization [equal], Formal analysis [equal], Methodology [equal], Writing—original draft [lead]), Kristoffer S. Berlin (Conceptualization [equal], Formal analysis [lead], Methodology [equal], Supervision [equal], Validation [lead], Writing—original draft [equal], Writing—review and editing [equal]), Jennifer Longoria (Data curation [equal], Methodology [equal], Project administration [equal], Writing—review and editing [equal]), Brian Potter (Data curation [supporting], Writing—review and editing [supporting]), Darcy Raches (Data curation [supporting], Writing—review and editing [supporting]), Jane Hankins (Funding acquisition [lead], Methodology [supporting], Project administration [supporting], Writing—review and editing [supporting]), Clifford Takemoto (Funding acquisition [supporting], Project administration [supporting], Writing—review and editing [supporting]), and Andrew Heitzer (Conceptualization [equal], Formal analysis [supporting], Methodology [lead], Supervision [lead], Writing—review and editing [lead])

Funding

This work was supported by the American Lebanese Syrian Associated Charities (ALSAC). Andrew M. Heitzer was supported by K23HL166697 (National Heart, Lung, and Blood Institute) during the time of this study.

Conflicts of interest

Andrew M. Heitzer received consultancy fees from Global Blood Therapeutics.