Materials

The CSHQ-A ( ^{NICHD SECCYD-Wisconsin, 2017} ) was used in this study to assess sleep problems. These questionnaires were attained primarily by a retrospective method with the parents/caregivers recalling on last week sleep patterns, disturbances, or behaviors (e.g., bed time, sleep behavior, waking during the night, morning wake up) ( ^{NICHD SECCYD-Wisconsin, 2017} ). The 22 items of the CSHQ-A are rated on a five-point scale ranging from 4 = “always” if the sleep behavior occurred seven times in the past week, 3 = “usually” if the behavior occurred five - six times in the past week, 2 = “sometimes” for two to four times that week, 1 = “rarely” for once a week, and 0 = “never” for zero times that week. A total score of more than 30 on CSHQ-A was taken as abnormal and indicative of sleep problems ( ^{NICHD SECCYD-Wisconsin, 2017} ). The questionnaire CSHQ-A ( ^{NICHD SECCYD-Wisconsin, 2017} ) was translated into Spanish with doctor Judith Owens written permission; the English back translation was revised afterwards by a native translator. This questionnaire is under its validity process in Colombia.

Procedures

Patients were ascertained in a University Hospital at Medellin, Colombia. An experienced child psychiatrist interview both parents (caregivers) and patients in order to make the ADHD diagnosis according to the DSM-5 and also psychiatric comorbidities. Data was collected regarding demographic characteristics (subject and family), ADHD presentation either combined (ADHD-C) or inattentive (ADHD-I), psychiatric comorbidities, and current medication (last two months or longer). Clinical medical records were reviewed in order to verify possible exclusion criteria. Parents (or caregivers) were asked to fill out the CSHQ-A and then it was verified to avoid unchecked answers.

The following comorbidity groups were determined: disruptive disorders (oppositional and defiant, intermittent and explosive, and conduct disorders), anxiety disorders (separation anxiety, generalized anxiety, social anxiety, obsessive compulsive, posttraumatic disorders), major depression disorder (MDD), learning, language and autism spectrum disorders (specific learning, language and autism spectrum disorders), bipolar spectrum disorders (bipolar and unspecified bipolar disorder), other disorders (low frequency comorbidities were group together such as tics disorders, feeding disorders and enuresis), and without comorbidities (ADHD pure).

Medication groups were determined as follows: stimulants (immediate released methylphenidate, long-acting methylphenidate and OROS), selective norepinephrine reuptake inhibitors (SNRI, such as atomoxetine), selective serotonin reuptake inhibitors (SSRI, such as fluoxetine, sertraline, fluvoxamine), second generation antipsychotics (SGA, such as risperidone, quetiapine, aripiprazole, olanzapine), alfa-adrenergic (clonidine), melatonin, and without medication (medication naive or no medication during the last two months). If a patient was not currently taking a prescribed treatment for any reason, he/she was classified as without treatment in this study. A variable called “medication that can cause insomnia” was arranged to divide the group in those who took stimulants and those who did not, and another variable called “medication that can cause somnolence” to divide the group into those who took any of these medications: SGA, clonidine, and/or melatonin, and those who did not. In this study, only medications that patients were taking before filling out the questionnaire were described. Although a significant percentage of patients had prescribed medications in prior consultations, due to difficulties in the Colombian health system they were not taking them at the time of the study, so they were classified as without treatment. This study was approved by the Ethics Committee Faculty of Medicine, University of Antioquia and Hospital San Vicente Fundación.

Statistical methods

Participants were classified according to whether they presented ADHD comorbidity and/or prescribed medication. They were described according to their demographic and clinical characteristics, and also by the CSHQ-A result, using median and interquartile range (25th and 75th percentile) in the quantitative variables and frequency and percentage in the categorical variables. Groups were compared using Chi-square test or Fisher exact test, Mann-Whitney U test, and effect size ( ^{Austin, 2009} ; ^{Fritz, Morris, & Richier, 2012} ). For non-parametric effect size we assumed a high effect size above .5, moderate between .3 and .5, and low under .3, and a level of significance p < .05 ( ^{Coolican, 2009} ; ^{Fritz et al., 2012} ). Using a linear regression model, the difference was estimated in adjusted means with their respective 95% confidence interval (CI) between groups: ADHD with comorbidity versus ADHD pure, and between ADHD with treatment and ADHD without treatment. Those variables that showed significant differences according to the Hosmer-Lemeshow criterion (p < .25), were adjusted ( ^{Hosmer & Lemeshow, 2000} ). These analyzes were performed in IBM SPSS version 23, and STATA version 14.0.

Sleep habits profile in ADHD

Latent class analysis (LCA) was used to explore the sleep habits profile according to ADHD presentation, demographic characteristics, comorbidity and/or prescribed medication ( ^{Lanza & Rhoades, 2013} ). This type of analysis allows to include both quantitative and qualitative variables. In this analysis the following were included: age (children/adolescents), gender (male/female), socioeconomic status (low/high), parental age, separation (yes/no), ADHD presentation (ADHD-C/ADHD-I), comorbidity (presence/absence), medication (yes/no), and the CSHQ-A total score. This analysis proposes that the sample can be divided into K number of classes. The Latent Gold 3.0 software ( ^{Vermunt & Magidson, 2000} ) was used to determine the optimal number of classes (profiles). Models were estimated from one class (under the assumption of homogeneity) up to 10 classes (under the assumption of heterogeneity in the sleep habits profiles of K latent class in ADHD). For the selection of the model with K classes, the Bayesian Information Criterion (BIC) was used, so it was selected the best fitting model with the lowest BIC among 10 models ( ^{Vermunt & Magidson, 2002} ). Here, the conditional probabilities that indicate the chance of having the characteristic according to class membership are shown, or otherwise the equally conditional mean in the quantitative variables. In the same way, the prevalence of each class is shown and for the chosen model the percentage of classification errors and the entropy (pseudo R² that indicates the best fitting model: the closer to one is the better). Entropy refers to how well subjects have been classified into each latent group.

Results of the CSHQ-A

Questionnaires were filled out mostly by mothers (87.3%), and the rest were answered by fathers or other caregivers. There were no missing data. Taking into account as sleep problems only those answers with higher scores (“always” = 4 points), this sample of ADHD presented: “child is restless during sleep” (27.6%), “child resists going to bed at bedtime” (21.9%), night awakenings “child wakes up at once during the night” (16.2%) and waking too early (18.9%). As for “child falls asleep within 20 minutes after going to bed” (“never” = 4 points), 8.3% of the sample had sleep onset delay. The analysis of the CSHQ-A total score and every dimension by groups according to the presence of comorbidity or medication is shown in Table 3.

The ADHD group with comorbidity had higher scores in the CSHQ-A than the ADHD pure group, especially in the dimension of hypersomnia. After adjusting for gender, socioeconomic status, fathers age, mothers age, ADHD presentation, and comorbidity, the difference is only one point (95% CI = -4, 2). Although hypersomnia item had still a significant difference, the effect size was low. The ADHD group with treatment had a higher score in the CSHQ-A than the group without treatment, not only for total score (median: 25 versus 18 points), but also for each dimension studied. After adjusting by gender, age, ADHD presentation, socioeconomic level, age of the father, and age of the mother and comorbidity, the difference was five points (95% CI = 2,8). In all dimensions there were significant differences with moderate to low effect sizes, except sleep duration. (Table 3)

The latent classes analysis (LCA) allowed to identify four groups. Information regarding the test criteria to find the best number of latent classes can be found in Table 4. The model of four classes for the sleep profile in ADHD was selected, with these properties: Log-likelihood -3563.6682, BIC: 7681.1297, number of parameters in model 102, classification errors .0399, and entropy of .9190. Table 5 presents the conditional probability for each characteristic according to de class, and Figure 1 presents the frequency of comorbidity in each class. Class 1 (34%) only children, males with young parents, most with ADHD-C (57%), half with comorbid ADHD (especially disruptive learning or language disorders) many with treatment (41%), and the CSHQ-A showed many sleep problems (mean: 25.98).Class 2: mostly adolescents (64%), female, ADHD-I (66%), comorbid ADHD (especially with anxiety disorder) without treatment and did no show sleep problems (mean: 12.78). Class 3: only adolescents, males, ADHD-C (64%), comorbid ADHD (66%), (especially with MDD), half with treatment, and CSHQ-A showed many sleep problems. Class 4: only children, males, half of the ADHD-C group, the majority with comorbid ADHD (especially disruptive learning or language disorders), older parents (not separated), the majority with treatment (60%), and CSHQ-A showed some sleep problems (Table 5). Regarding the presence of comorbidity with anxiety disorders or MDD there were statistically significant differences for class 2 and class 3 (p < .001 and p = .007 respectively) (Figure 1).

Figure 1. Frequency of comorbidities according to the 4 classes in ADHD.  

Limitations

The CSHQ-A is under the process of validation in Colombia, and there are still no reference scores in the population. The instrument is used worldwide in children between four and 10 years of age but to a lesser extent in adolescents. Adolescents may have different sleep problems than children, and it has to be considered that parents identify less adolescents’ sleep patterns. For this reason, for further studies we suggest to use a scale of adolescent sleep self-report in addition to the parent questionnaire to achieve a better description.

Some study limitations regarding the LCA are worth mentioning. First, the number of parameters of the resulting model was 102, that means 2.23 participants per parameter. Although the required sample size depends on several aspects, rule of thumb is at least five participants per parameter ( ^{Gorsuch, 1983} ), so it is required to reproduce the model in a larger sample. Second, other classes could be obtained if the CSHQ-A dimensions are used instead of the total score. Finally, instead of using the comorbidity variable (presence/absence), each comorbidity could be included in the latent model class. The same is also true for the case of medications. This was not done because it would increase the number of model parameters. Therefore, additional studies with larger sample sizes are required. Finally, these results cannot be generalized to other populations.

In this sample of children and adolescents with ADHD, the ADHD-C presentation was associated with a higher frequency of comorbidity. These subjects were more likely to be in psychopharmacological treatment and have a greater report of sleep problems by the CSHQ-A in most dimensions compared to those who do not receive treatment. Subjects without comorbidity had fewer sleep problems than those with comorbid ADHD. Through the latent classes model, ADHD groups that present more sleep problems were identified.