Success of randomization/tests for confounding
Children who saw trained clinicians were more likely to be female (49 vs. 39%, p=.02) and slightly older (10.7 vs. 10.1 years, p=.08) (see ). Otherwise there was no difference between the study groups in SDQ classification, other demographics, insurance, prior mental health counseling or co-morbidity, or chronic disease ICD –9 codes. Parents of children seeing control and trained clinicians did not differ by age or educational level.
Baseline comparison of study groups including children with primary care visits in the six month follow-up period (adjusted for clustering by clinician); n=397 children.
Children whose index visit was with a trained clinician saw a trained clinician for 67% of their subsequent primary care visits, whereas children whose index visit was with a control clinician saw a control or other untrained primary care clinician for 92% of visits.
Because of the slight age and sex imbalances at baseline between the intervention and control groups, the possibility of confounding was evaluated. There was no relationship between age and SDQ status (p=0.16) or age and the number of primary care visits (p=0.59). In addition, there was no relationship between the child’s sex and the number of primary care visits (p=0.99).
Utilization by SDQ status
Utilization of primary care varied by baseline SDQ status (). Compared to children classified as SDQ unlikely (but nominated by clinicians as having problems), children classified as SDQ possible made, on average, 0.38 more primary care visits (p=0.01), while children classified as SDQ probable made 0.65 more such visits during the six-month follow-up period (p < 0.0001, with overall p-value of 0.0002). The standard errors of the means are small, i.e. about 1/20th of the mean, suggesting little dispersion and implying that many patients had just one more visit beyond the index as opposed to a few patients coming in more frequently. Specifically, for SDQ unlikely, the range of visits was 1–6 with 73% of observations being a 1 or 2 (mean = 2.1 vs median = 2.0). For SDQ possible, the range of visits was 1–9 with 56% of observations being a 1 or 2 (mean = 2.47 vs median = 2.0). For SDQ probable, the range of visits was 1–9 with 57% of observations being a 1 or 2 (mean = 2.75 vs median = 2.0).
Mean number of primary care visits by SDQ status and study group (adjusted for clustering by clinician) with no significant interaction between SDQ status and study group (p=0.26); n =397 children.
Over the six month follow-up period, clinician-documented psychotropic medication prescription also varied significantly by SDQ status. Among children classified as SDQ unlikely, 11% had documentation of one or more psychotrophic medication prescriptions compared to 14% in the SDQ possible and 34% in the SDQ probable groups (Rao-Scott p < 0.0001). Similarly, there was a significant relationship between SDQ status and receipt of counseling before the index visit (p<0.0001). Psychiatric referral did not vary significantly by SDQ status, however the percentage of children referred was low (1 % SDQ unlikely, 3.6% SDQ possible, 5.8% SDQ probable, Rao- Scott p = 0.16). There was no relationship between SDQ status and having any chronic disease (p = 0.30).
Therefore, receipt of prior counseling was added to the model predicting primary care visits. When prior counseling was added to this model, there was no relationship between the number of primary care visits and prior counseling (p=0.84), nor was there a relationship with an interaction term for SDQ status and prior counseling (p=0.84). SDQ status remained significantly associated with the number of primary care visit in this model (p=0.008).
Impact of training on primary care utilization
The mean number of visits for children seeing trained clinicians was nearly identical to the mean among children seeing control clinicians (2.46 vs. 2.52) (). shows the relationship of clinician training status to the mean number of visits per child for various types of visits. There were no statistically significant differences for any of the four levels of office visits (ranging from brief through 40 minutes), or for health maintenance visits (of children 5–11 and 12–17). There were no statistically significant differences in the proportion of children/youth who had visits for asthma, allergic rhinitis, or diabetes. There was a trend for trained clinicians to prescribe fewer psychotropic medications than control clinicians. Trained clinicians prescribed one or more psychotropic medications to 17% of the children they saw compared to 26% in the control group (Rao-Scott p =0.11). Trained clinicians documented psychiatric referral for 3.7% of the children they saw compared to 3.9% in the control group (Rao-Scott p =0.93).
Comparison, by study group, of mean number of primary care visits stratified by CPT code, well child visits and visits for select chronic diseases (adjusted for clustering by clinician); n =397 children.
Subset analysis of all visit types
A subset analysis of all visits occurring in the integrated rural health network showed similar results when all visit types are included, although, as would be expected, the mean number of visits was higher. Of the 7 clinical sites included, there were 21 clinicians, 188 children and 593 visits of all types during the study period. The mean number of all visits by treatment group was not significantly different, i.e. 3.57 intervention versus 3.38 for control, p = 0.605, adjusted for clustering by clinician. However, the mean number of all visits did differ by SDQ status as follows: 2.98 for SDQ unlikely, 3.71 for SDQ possible and 3.64 for SDQ probable (p = 0.026). Further stratification by visit type was not possible due to sample size limitations.