OBJECTIVES:

For pediatric complex care programs to target enhanced care coordination services to the highest-risk patients, it is critical to accurately identify children with medical complexity (CMC); however, no gold standard definition exists. The aim of this study is to describe a point-of-care screening algorithm to identify CMC with high health care use, a group that may benefit the most from improved care coordination.

METHODS:

From July 1, 2015, to June 30, 2016 (fiscal year 2016 [FY16]), a medical complexity screening algorithm was implemented by a pediatric complex care program at a single tertiary care center for hospitalized patients at the time of admission. Using the screening algorithm, we categorized inpatients into 1 of 3 groups: CMC, children with special health care needs (CSHCN), or previously healthy (PH) children. Inpatient resource use for FY16 and FY17 encounters was extracted for children screened in FY16.

RESULTS:

We categorized 2187 inpatients in FY16 into the 3 complexity groups (CMC = 77; CSHCN = 1437; PH children = 673). CMC had more complex chronic conditions (median = 6; interquartile range [IQR] 4–11) than CSHCN (median = 1; IQR 0–2) and PH children (median = 0; IQR 0–0). CMC had greater per-patient and per-encounter hospital use than CSHCN and PH children. CMC and children with ≥4 complex chronic conditions had comparable levels of resource use.

CONCLUSIONS:

By implementation of a point-of-care screening algorithm, we identified CMC with high health care use. By using this algorithm, it was feasible to identify hospitalized CMC that could benefit from care coordination by a pediatric complex care program.

Children with medical complexity (CMC) are a complex and costly patient population. Although they represent <1% of the general pediatric population, they account for nearly 40% of total child health expenditures, up to 50% of inpatient pediatric costs, and as much as 70% of unplanned 30-day hospital readmissions.1,2  Additionally, CMC are a clinically heterogeneous group of children characterized by a wide range of multiple chronic conditions, a need for technology assistance (TA), a need for multiple subspecialists, and high resource use. Given their high resource use, risk for poor outcomes,3,4  and need for higher value care,57  tertiary care centers nationwide have established pediatric complex care programs.

Pediatric complex care programs employ differing strategies for the critical task of identifying CMC. A systematic approach to identify CMC can help complex care programs focus their efforts on CMC who are objectively identified by their level of medical complexity. Identifying these children is important, given the limited resources available and potentially large numbers of CMC in each program’s referral region. Although identification of a CMC may seem evident during individual clinical encounters, identification of CMC at the population level is challenging.811 

Population level, evidence-based approaches, such as the Pediatric Medical Complexity Algorithm (PMCA) or complex chronic condition (CCC) codes, can be used to accurately identify CMC.8,10,11  Limitations of these existing systems include (1) a reliance on retrospective analysis of diagnosis codes from administrative databases; (2) an inability to identify CMC who will have persistent high resource use; and (3) limited utility at the point of care. Systematic point-of-care approaches to identify CMC that are integrated into clinical practice could be used to address the limitations of existing medical complexity classification systems; however, none in the literature have been previously described.

The aim of this study was to describe a point-of-care complexity screening algorithm to identify CMC with high health care use. We hypothesized that CMC identified by using our screening algorithm would have higher use measures both per encounter and per year compared with children that did not meet this definition. As a secondary aim, we compared health care use between CMC identified by using our screening algorithm and those identified by using CCC codes.

This study was conducted at Duke Children’s Hospital, a 190-bed pediatric tertiary care center located in Durham, NC, with >6000 hospital admissions per year. Duke Children’s Complex Care Service (CCS) is an interdisciplinary complex care program that cares for a regional CMC population.12  Enrollment into the CCS program occurs during an inpatient admission; subsequently, patients receive augmented care coordination across the inpatient-outpatient continuum. CMC with high health care use who are eligible for program enrollment are identified through review of the inpatient general pediatric and surgical subspecialty daily census and application of an internally developed point-of-care complexity screening algorithm. From July 1, 2015, to June 30, 2016 (fiscal year 2016 [FY16]), the CCS program coordinator (V.J.) applied the complexity screening algorithm via manual chart review to all patients admitted to the general pediatric or surgical teams at Duke Children’s Hospital and categorized inpatients into 1 of 3 groups: previously healthy (PH), children with special health care needs (CSHCN), or CMC. A secure online administrative database (Research Electronic Data Capture13 ) was used to track of the number of patients screened and time spent screening per day.

If the same patient was hospitalized more than once and at least 6 months had passed since the original screening date, complexity screening was repeated and logged as a separate unique screen. An independent manual chart audit was performed by 3 attending physicians (D.M., L.P., and V.P.) to confirm the appropriate complexity level for a sample of patients whose complexity classifications at different admissions were discordant (ie, complexity level decreased or increased across 2 screenings, eg, CMC during first admission and then CSHCN at subsequent admission).

The point-of-care medical complexity algorithm was based exclusively on electronic health record chart review and included 2 steps. In the first step, we distinguished CSHCN from PH children using the Children with Special Health Care Needs Screener: a validated 5-item yes or no questionnaire that is used to assess for the presence of chronic conditions expected to last >12 months that lead to functional limitations or require prescription medications, health services, special therapies, and/or behavioral health treatment or counseling.14  Our program coordinator applied the Children with Special Health Care Needs Screener’s scoring rubric (the presence of ≥1 affirmative responses to the 5 core items defined a child as CSHCN) in a novel way. Instead of gathering responses to the Children with Special Health Care Needs Screener directly from parents, questions were answered on the basis of information available in the child’s electronic medical record, such as medication prescriptions, chronic conditions, and provider documentation. For example, a child was given a yes response to the question “does the child need or get special therapy, such as physical, occupation, or speech therapy” only if there was a documented therapy visit in the child’s chart in the past 12 months.

The second step was used to identify CMC as a subset of CSHCN by incorporating supplemental criteria to the Children with Special Health Care Needs Screener. All 4 of the following supplemental criteria must be met for a child to be categorized as a CMC: (1) high resource use (≥2 inpatient admissions, ≥1 ICU admissions, or ≥6 ED visits in the preceding 12 months), (2) long-term medical TA (limited to feeding tube, ventriculoperitoneal shunt, tracheostomy, central line, or home mechanical ventilation), (3) seen by ≥3 subspecialists as an outpatient within our health care system in the preceding 12 months, and (4) ≥4 positive questions on the Children with Special Health Care Needs Screener.

To compare our point-of-care categorization of CMC to retrospective approaches described in the literature, we retrospectively classified children using the CCCs classification system.15  Similar to Cohen et al,1  we divided children with any CCC into those with neurologic impairment (NI), those with multiple CCCs, and those with a single CCC. To ascertain TA, we extracted all diagnostic and procedural International Classification of Diseases, Ninth Revision, and International Classification of Diseases, 10th Revision, codes during inpatient encounters to identify insertions and removals of medical devices.1  We used diagnostic codes from ∼4 years of historic encounters to identify CCC codes.

Internal health resource use data and demographics were derived from our institution’s administrative database and included all hospital encounters from July 1, 2015, to June 30, 2016 (FY16), and from July 1, 2016, to June 30, 2017 (fiscal year 2017 [FY17]), for the cohort of children screened during FY16. Costs included total inpatient costs, inclusive of fixed (equipment, utilities, building, etc) and variable (staffing, materials, etc) costs. Professional charges were not included in costs. We reported use measures both at the patient and encounter level. Per-encounter measures included administrative data from a single encounter. Considering that an individual patient may have had multiple encounters during the study period, per-patient-per-month measures represented the sum of all encounters’ use measures from screening to the end of fiscal year 2017 divided by 1 month (30 days).

Statistical analyses were performed by using SAS software, version 9.4, for the SAS System for Linux (SAS Institute, Inc, Cary, NC). Descriptive demographic and patient characteristic data were compiled from the raw data and summarized by medical complexity group, with counts and percentages for categorical items and median and 25th and 75th percentiles for continuous items. For statistical analyses for differences in patient characteristics and demographics by medical complexity, we included the Kruskal-Wallis test for continuous variables and Fisher’s exact test for categorical variables (except for race, in which χ2 testing was used because of computational limitations). Step-down pairwise tests were conducted by using the Wilcoxon rank test for continuous variables and Fisher’s exact test for all categorical variables. Multiple linear regression (continuous measures) or logistic regression (dichotomous measures) models were developed to assess the relationship between complexity level and resource use. Age, race, payer, and Hispanic ethnicity were included in all models. The Duke Institutional Review Board approved this investigation.

In FY16, we screened and categorized 2187 children into 3 complexity groups: 77 CMC (3.5%), 1437 CSHCN (65.7%), and 673 PH children (30.8%). The 3 medical complexity groups differed by several demographic and medical characteristics (Table 1). First, CSHCN were older than either CMC or PH children (median of 8.91 vs 5.91 or 2.79 years old, respectively; P values .002 and <.001, respectively), and CMC were older than PH children (5.91 vs 2.91 years old; P value .002). Second, PH children were more likely to be Hispanic than CSHCN (14% vs 9%; P value .001), and an overall comparison among groups revealed significant differences in race (P value .013). Third, ∼1 in 5 CSHCN required TA (21.1%; n = 303), with the most common being enteral feeding tubes (53%; 161 of 303). By using the screening criteria definition, all CMC required TA (100.0%; 77 of 77), whereas no PH children required it (0.0%; 0 of 673). Enteral feeding tubes were the most common technology used by CMC (85.7%). Fourth, we found statistically significant differences in the outpatient subspecialty care and number of CCC codes by medical complexity. CMC had seen a median of 5 different subspecialists (interquartile range [IQR] 4–6) and had 6 CCC qualifying codes (IQR 4–11), compared with 2 different subspecialists (IQR 1–2) and 1 CCC qualifying code (IQR 0–2) for CSHCN and 0 subspecialists (IQR 0–0) and 0 CCC qualifying codes (IQR 0–0) for PH children (all pairwise P values <.001). Lastly, 84% of CMC had NI, as defined by International Classification of Diseases diagnostic codes, compared with 38% of CSHCN and 7% of PH children (CMC versus CSHCN P value <.001; CMC versus PH children P value <.001; Fig 1)

TABLE 1

Demographics and Patient Characteristics at Time of Admission by Medical Complexity for Children Screened in Fiscal Year 2016

CMC (n = 77)CSHCN (n = 1437)PH (n = 673)
Age, y, median (25th, 75th percentile) 6.0 (3.30, 9.93) 8.9 (3.09, 14.78) 2.8 (0.86, 11.53) 
Sex, female, n (%) 33 (42.9) 669 (46.6) 285 (42.3) 
Race, n (%)    
 Black or African American 23 of 76 (30.3) 444 of 1392 (31.9) 223 of 639 (34.9) 
 White 42 of 76 (55.3) 745 of 1392 (53.5 289 of 639 (45.2) 
 ≥2 races 5 of 76 (6.6) 83 of 1392 (6.0) 37 of 639 (5.8) 
 Asian American 3 of 76 (3.9) 32 of 1392 (2.3) 26 of 639 (4.1) 
 Other 6 of 76 (7.9) 120 of 1392 (8.6) 90 of 639 (14.1) 
Ethnicity, Hispanic, n (%) 7 (9.1) 121 of 1385 (8.7) 89 of 639 (13.9) 
Primary payer, n (%)    
 Medicaid or Medicare 42 (54.5) 786 (54.7) 398 (59.1) 
 Private or other 35 (45.5) 651 (45.3) 275 (40.9) 
Technology on admission (by using chart review), n (%)    
 Ventriculoperitoneal shunt 27 (35.1) 124 (8.6) 0 (0) 
 Enteral feeding tube 66 (85.7) 161 (11.2) 0 (0) 
 Central venous catheter 11 (14.3) 58 (4.0) 0 (0) 
 Tracheostomy 17 (22.1) 13 (0.9) 0 (0) 
 Ventilator 14 (18.2) 19 (1.3) 0 (0) 
 Any of the above technology 77 (100) 303 (21.1) 0 (0) 
No. subspecialists seen in the last 12 mo    
 Median (25th, 75th percentile) 5 (4, 6) 2 (1, 2) 0 (0, 0) 
No. questions positive on the Children with Special Health Care Needs Screener    
 Median (25th, 75th percentile) 5 (4, 5) 3 (3, 4) 0 (0, 0) 
High user in the last 12 mo, n (%)a 73 (94.8) 549 (38.2) 4 (0.6) 
Measures of complexity by using diagnostic codes    
 No. unique CCC qualifying codes, median (25th, 75th percentile) 6 (4, 11) 1 (0, 2) 0 (0, 0) 
 0 CCC, n (%) 1 (1.3) 675 (47.0) 571 (84.8) 
 1–3 CCC, n (%) 14 (18.1) 569 (39.6) 93 (13.8) 
 ≥4 CCC, n (%) 62 (80.5) 193 (13.4) 9 (1.3) 
 NIb, n (%) 65 (84.4) 549 (38.2) 49 (7.3) 
 TA (by using diagnostic codes)a, n (%) 75 (97.4) 417 (29.0) 19 (2.8) 
CMC (n = 77)CSHCN (n = 1437)PH (n = 673)
Age, y, median (25th, 75th percentile) 6.0 (3.30, 9.93) 8.9 (3.09, 14.78) 2.8 (0.86, 11.53) 
Sex, female, n (%) 33 (42.9) 669 (46.6) 285 (42.3) 
Race, n (%)    
 Black or African American 23 of 76 (30.3) 444 of 1392 (31.9) 223 of 639 (34.9) 
 White 42 of 76 (55.3) 745 of 1392 (53.5 289 of 639 (45.2) 
 ≥2 races 5 of 76 (6.6) 83 of 1392 (6.0) 37 of 639 (5.8) 
 Asian American 3 of 76 (3.9) 32 of 1392 (2.3) 26 of 639 (4.1) 
 Other 6 of 76 (7.9) 120 of 1392 (8.6) 90 of 639 (14.1) 
Ethnicity, Hispanic, n (%) 7 (9.1) 121 of 1385 (8.7) 89 of 639 (13.9) 
Primary payer, n (%)    
 Medicaid or Medicare 42 (54.5) 786 (54.7) 398 (59.1) 
 Private or other 35 (45.5) 651 (45.3) 275 (40.9) 
Technology on admission (by using chart review), n (%)    
 Ventriculoperitoneal shunt 27 (35.1) 124 (8.6) 0 (0) 
 Enteral feeding tube 66 (85.7) 161 (11.2) 0 (0) 
 Central venous catheter 11 (14.3) 58 (4.0) 0 (0) 
 Tracheostomy 17 (22.1) 13 (0.9) 0 (0) 
 Ventilator 14 (18.2) 19 (1.3) 0 (0) 
 Any of the above technology 77 (100) 303 (21.1) 0 (0) 
No. subspecialists seen in the last 12 mo    
 Median (25th, 75th percentile) 5 (4, 6) 2 (1, 2) 0 (0, 0) 
No. questions positive on the Children with Special Health Care Needs Screener    
 Median (25th, 75th percentile) 5 (4, 5) 3 (3, 4) 0 (0, 0) 
High user in the last 12 mo, n (%)a 73 (94.8) 549 (38.2) 4 (0.6) 
Measures of complexity by using diagnostic codes    
 No. unique CCC qualifying codes, median (25th, 75th percentile) 6 (4, 11) 1 (0, 2) 0 (0, 0) 
 0 CCC, n (%) 1 (1.3) 675 (47.0) 571 (84.8) 
 1–3 CCC, n (%) 14 (18.1) 569 (39.6) 93 (13.8) 
 ≥4 CCC, n (%) 62 (80.5) 193 (13.4) 9 (1.3) 
 NIb, n (%) 65 (84.4) 549 (38.2) 49 (7.3) 
 TA (by using diagnostic codes)a, n (%) 75 (97.4) 417 (29.0) 19 (2.8) 

Ventilator includes traditional ventilator or bilevel positive airway pressure.

a

High user was defined as ≥2 inpatient admissions, ≥1 ICU admissions, or ≥6 ED visits in the preceding 12 mo; 4 of 77 patients in the CMC group did not meet high user criterion but were defined as CMC and enrolled into CCS after clinician case discussion.

b

NI and TA defined by using criteria from Cohen et al1 .

FIGURE 1

CCC screening algorithm and FY2016 screening results. a No response to any 1 of these questions classified the child as CSHCN. b Medical technology was defined as a long-term need for any 1 of the following: (1) ventriculoperitoneal shunt, (2) long-term enteral feeding tube (eg, gastrostomy tube, jejunostomy tube, gastro-jejunal tube, etc), (3) long-term central venous catheter, (4) tracheostomy, (5) mechanical ventilation-dependent (includes traditional ventilator or bilevel positive airway pressure). Other equipment (such as supplemental oxygen, continuous positive airway pressure, mobility aids and devices, insulin needles and syringes, glucometers, vagus nerve stimulation, etc) is not sufficient to meet the medical equipment criteria. c High resource use was defined as yes to any of the following in the 12 months before screening: (1) ≥6 emergency department visits that did not result in an admission, (2) ≥2 inpatient admissions (includes ICU, stepdown, and floor admissions and excludes birth admission if <12 months old), and (3) ≥1 ICU admissions (includes any ICU time, on the basis of hospital course in discharge summary).

FIGURE 1

CCC screening algorithm and FY2016 screening results. a No response to any 1 of these questions classified the child as CSHCN. b Medical technology was defined as a long-term need for any 1 of the following: (1) ventriculoperitoneal shunt, (2) long-term enteral feeding tube (eg, gastrostomy tube, jejunostomy tube, gastro-jejunal tube, etc), (3) long-term central venous catheter, (4) tracheostomy, (5) mechanical ventilation-dependent (includes traditional ventilator or bilevel positive airway pressure). Other equipment (such as supplemental oxygen, continuous positive airway pressure, mobility aids and devices, insulin needles and syringes, glucometers, vagus nerve stimulation, etc) is not sufficient to meet the medical equipment criteria. c High resource use was defined as yes to any of the following in the 12 months before screening: (1) ≥6 emergency department visits that did not result in an admission, (2) ≥2 inpatient admissions (includes ICU, stepdown, and floor admissions and excludes birth admission if <12 months old), and (3) ≥1 ICU admissions (includes any ICU time, on the basis of hospital course in discharge summary).

Close modal

In FY16, of the 2187 children screened, the medical complexity classification of only 13 patients (0.5%) changed after a chart review of children with 2 discordant screenings; 3 were reclassified from PH to CSHCN, 8 from CSHCN to CMC, and 2 from CMC to CSHCN. The program coordinator screened an average of 11 patients per day in FY2016 and spent ∼5 to 10 minutes per patient screening (∼90 minutes per day). A total of 123 children (5.6%) were missing health care use data and were excluded from our use analyses, leaving a final sample size of 2064 patients (71 CMC, 1346 CSHCN, and 647 PH children). Of those screened in FY16, 28% (578 of 2064) had at least 1 inpatient encounter in FY17. Having an inpatient encounter varied by medical complexity; 70% CMC (50 of 71) vs 31% CSHCN (413 of 1346) vs 18% PH children (115 of 647) screened in FY16 had at least 1 subsequent encounter in FY17 (Table 2).

TABLE 2

Inpatient Resource Use in FY16 and FY17 of Children Screened in FY16 by Point-Of-Care Medical Complexity Screening

FY16–17
CMCCSHCNPHCMC Versus CSHCN, PaCSHCN Versus PH, Pa
Per-encounter use measures n = 782 n = 4663 n = 1020 — — 
 Days hospitalized, median (IQR) 4 (2–8) 4 (2–7) 3 (2–5) 0.16 <0.01 
 Ever readmitted, n (%) 546 of 778 (70.2) 2092 of 4638 (45.1) 157 of 1014 (15.5) <0.01 <0.01 
 Readmitted within 14 d, n (%) 114 of 778 (14.7) 443 of 4638 (9.6) 63 of 1014 (6.2) <0.01 <0.01 
 Readmitted within 30 d, n (%) 197 of 778 (25.3) 698 of 4638 (15.0) 72 of 1014 (7.1) <0.01 <0.01 
 Readmitted within 90 d, n (%) 332 of 778 (42.7) 1177 of 4638 (25.4) 100 of 1014 (9.9) <0.01 <0.01 
 Total hospitalization cost, median (IQR) 6925 (3445–15 509) 6264 (3230–13 353) 3888 (1839–7790) 0.01 <0.01 
 Cost per hospitalized day, median (IQR) 1616 (1216–2280) 1534 (1058–2317) 1238 (812–1974) <0.01 <0.01 
 ICU stay, n (%) 311 (40.0) 1359 (29.3) 216 (21.4) <0.01 <0.01 
 Days in ICU, median (IQR) 3 (1–9) 2 (1–5) 2 (1–5) <0.01 0.92 
Per patient per month use measures n = 71 n = 1346 n = 647 — — 
 Days hospitalized per month, median (IQR) 0.82 (0.33–2.05) 0.21 (0.12–0.45) 0.12 (0.08–0.22) <0.01 <0.01 
 No. admissions per month, median (IQR) 0.12 (0.08–0.21) 0.04 (0.04–0.08) 0.04 (0.04–0.05) <0.01 <0.01 
 Total hospital charges per month, median (IQR) 1653 (724–4046) 404 (186–1078) 152 (72–313) <0.01 <0.01 
FY16–17
CMCCSHCNPHCMC Versus CSHCN, PaCSHCN Versus PH, Pa
Per-encounter use measures n = 782 n = 4663 n = 1020 — — 
 Days hospitalized, median (IQR) 4 (2–8) 4 (2–7) 3 (2–5) 0.16 <0.01 
 Ever readmitted, n (%) 546 of 778 (70.2) 2092 of 4638 (45.1) 157 of 1014 (15.5) <0.01 <0.01 
 Readmitted within 14 d, n (%) 114 of 778 (14.7) 443 of 4638 (9.6) 63 of 1014 (6.2) <0.01 <0.01 
 Readmitted within 30 d, n (%) 197 of 778 (25.3) 698 of 4638 (15.0) 72 of 1014 (7.1) <0.01 <0.01 
 Readmitted within 90 d, n (%) 332 of 778 (42.7) 1177 of 4638 (25.4) 100 of 1014 (9.9) <0.01 <0.01 
 Total hospitalization cost, median (IQR) 6925 (3445–15 509) 6264 (3230–13 353) 3888 (1839–7790) 0.01 <0.01 
 Cost per hospitalized day, median (IQR) 1616 (1216–2280) 1534 (1058–2317) 1238 (812–1974) <0.01 <0.01 
 ICU stay, n (%) 311 (40.0) 1359 (29.3) 216 (21.4) <0.01 <0.01 
 Days in ICU, median (IQR) 3 (1–9) 2 (1–5) 2 (1–5) <0.01 0.92 
Per patient per month use measures n = 71 n = 1346 n = 647 — — 
 Days hospitalized per month, median (IQR) 0.82 (0.33–2.05) 0.21 (0.12–0.45) 0.12 (0.08–0.22) <0.01 <0.01 
 No. admissions per month, median (IQR) 0.12 (0.08–0.21) 0.04 (0.04–0.08) 0.04 (0.04–0.05) <0.01 <0.01 
 Total hospital charges per month, median (IQR) 1653 (724–4046) 404 (186–1078) 152 (72–313) <0.01 <0.01 
a

P values were obtained from multiple regression (continuous measures) or multiple logistic regression (dichotomous measures). Analyses were adjusted for age, race, ethnicity, payer, and sex. —, not applicable.

In all FY16 per-encounter measures of use, with the exception of days hospitalized, CMC had higher use, compared with CSHCN and PH children (Table 2). In the subset of children with a FY17 hospitalization, CMC had persistently higher per-encounter use than CSHCN and PH children, including higher 30- and 90-day readmission rates, cost per hospital day, and percent with an ICU stay.

CMC also had higher per-patient-per-month use measures (including the total days hospitalized, number of admissions, and total hospital charges) in both FY16 and FY17, compared with CSHCN and PH children (Table 2). Figure 2 reveals the mean hospital charges per patient over the 2 years by our screening components and medical complexity designation. Although CMC represented 3% of hospitalized children admitted to the general pediatrics or surgical services in FY16, they accounted for 15% of inpatient costs in that year. In contrast, CSHCN represented 65% of the population and 71% of total costs in FY16, and PH represented 31% of the population and 15% of total costs in FY16.

FIGURE 2

Longitudinal hospital costs over 2 years by degree of medical complexity.

FIGURE 2

Longitudinal hospital costs over 2 years by degree of medical complexity.

Close modal

Comparison of the point-of-care screening algorithm to retrospective evidence-based classification systems revealed key similarities. CMC defined by using the screening algorithm had greater hospital use measures in comparison with children with ≥1 CCC code and were most similar to those with ≥4 CCC codes (Table 3). Approximately one-quarter of children with ≥4 CCC codes met the additional criteria in our screening algorithm to be defined as CMC. We did not find addition of other diagnostic code groupings, such as TA or NI, to a ≥4 CCC code definition of CMC significantly impacted health care use (data are not shown).

TABLE 3

Inpatient Resource Use in FY16 and FY17 Comparing CMC Identified by Point-of-Care Screening Algorithm to Diagnostic Code–Based Identification

FY16 and FY17
Point-of-Care Screening Algorithm Identification of CMCaDiagnostic Code–Based Identification of CMCa
Any CCC (≥1CCC)≥4 CCC
Per-encounter use measures    
 Encounters n = 277 n = 2096 n = 900 
 Days hospitalized, median (IQR) 4 (2–9) 4 (3–7) 5 (3–9) 
 Readmitted within 14 d, n (%) 45 of 273 (16.5) 235 of 2073 (11.3) 137 of 886 (15.5) 
 Readmitted within 30 d, n (%) 75 of 273 (27.5) 366 of 2073 (17.7) 215 of 886 (24.3) 
 Readmitted within 90 d, n (%) 120 of 273 (44.0) 569 of 2073 (27.4) 348 of 886 (39.3) 
 Total hospitalization cost, $, n (%) 7965 (4363–18 781) 7083 (3921–17 305) 7896 (4430–20 700) 
 Cost per hospitalized day, $, median (IQR) 1682 (1326–2457) 1671 (1206–2563) 1630 (1246–2323) 
 Any ICU stay, n (%) 118 (42.8) 689 (33.1) 316 (35.3) 
 Length of ICU stay, median (IQR) 3 (1–9) 2 (1–6) 3 (1–10) 
Per patient per mo use measures    
 Patients n = 71 n = 874 n = 236 
 Total days hospitalized per mo, median (IQR) 0.82 (0.33–2.05) 0.29 (0.16–0.86) 0.86 (0.29–2.40) 
 No. admissions per mo, Median (IQR) 0.12 (0.08–0.21) 0.08 (0.04, 0.12) 0.12 (0.08, 0.21) 
 Total hospital charges per mo, median (IQR) 40 278 (17 652–98 589) 17 343 (7216–43 757) 39 680 (13 594–127 984) 
 No. and percent of identified patients also in CMC column, n (%) — 69 (7.9) 55 (23.5) 
FY16 and FY17
Point-of-Care Screening Algorithm Identification of CMCaDiagnostic Code–Based Identification of CMCa
Any CCC (≥1CCC)≥4 CCC
Per-encounter use measures    
 Encounters n = 277 n = 2096 n = 900 
 Days hospitalized, median (IQR) 4 (2–9) 4 (3–7) 5 (3–9) 
 Readmitted within 14 d, n (%) 45 of 273 (16.5) 235 of 2073 (11.3) 137 of 886 (15.5) 
 Readmitted within 30 d, n (%) 75 of 273 (27.5) 366 of 2073 (17.7) 215 of 886 (24.3) 
 Readmitted within 90 d, n (%) 120 of 273 (44.0) 569 of 2073 (27.4) 348 of 886 (39.3) 
 Total hospitalization cost, $, n (%) 7965 (4363–18 781) 7083 (3921–17 305) 7896 (4430–20 700) 
 Cost per hospitalized day, $, median (IQR) 1682 (1326–2457) 1671 (1206–2563) 1630 (1246–2323) 
 Any ICU stay, n (%) 118 (42.8) 689 (33.1) 316 (35.3) 
 Length of ICU stay, median (IQR) 3 (1–9) 2 (1–6) 3 (1–10) 
Per patient per mo use measures    
 Patients n = 71 n = 874 n = 236 
 Total days hospitalized per mo, median (IQR) 0.82 (0.33–2.05) 0.29 (0.16–0.86) 0.86 (0.29–2.40) 
 No. admissions per mo, Median (IQR) 0.12 (0.08–0.21) 0.08 (0.04, 0.12) 0.12 (0.08, 0.21) 
 Total hospital charges per mo, median (IQR) 40 278 (17 652–98 589) 17 343 (7216–43 757) 39 680 (13 594–127 984) 
 No. and percent of identified patients also in CMC column, n (%) — 69 (7.9) 55 (23.5) 

—, not applicable.

We compared CMC with ≥4 CCC codes with both (1) CMC with <4 CCC codes and (2) CSHCN with ≥4 CCC codes (Supplemental Table 4). We found that CMC with <4 CCC codes were less likely to have TA by using diagnostic codes compared with CMC with ≥4 CCC codes (87% vs 100%; P value .036). However, 100% of all CMC (<4 and ≥4 CCC codes) had TA, by using point-of-care screening. Analysis of CMC with ≥4 CCC codes compared with CSHCN with ≥4 CCC codes revealed that CMC were more likely to require TA (100% vs 57% by screenings; P value <.001), see more subspecialists (5 vs 2; P value <.001), and have NI (84% vs 50%; P value <.001).

With the point-of-care screening algorithm described in this study, we stratified an inpatient pediatric population by medical complexity level and identified CMC with high resource use. In contrast to previous literature, in which researchers largely describe the retrospective application of diagnostic codes to identify CMC,8,10  our algorithm was applied at the point of care during an inpatient clinical encounter. Application of the algorithm during a hospitalization offered an efficient approach to identify CMC who may benefit from enhanced care coordination via a structured complex care program.

In this study, CMC had higher encounter-level and patient-level use and accounted for a disproportionate amount of inpatient costs, compared with CSHCN and PH children. However, when compared to previous studies, we anticipated CMC would represent an even greater proportion of inpatient costs; instead, CMC accounted for 15% of inpatient costs included in our study, compared to as much as 40% of inpatient costs in the literature.16,17  The lower proportion of inpatient costs observed in our study was likely due to the exclusion of children admitted to several subspecialty services (including cardiology, hematology and oncology, endocrinology, and nephrology) and a more restrictive definition of CMC; we suspect many children screened as CSHCN by using our definition would be classified as CMC by using alternative approaches such as the PMCA or CCC approaches. For example, within this sample of children, 891 of them would have been classified as CMC if using at least 1 CCC as the criterion, and 246 would have been classified as CMC for ≥4 CCCs; yet, only 74 were classified as CMC by using our algorithm, likely because of the inclusion of additional CMC criteria beyond diagnosis codes alone (eg, multispecialty care, high past hospital use, and TA). We did not validate our screening algorithm’s ability to identify CMC because no gold standard approach to identify CMC at the point of care exists.

In contrast to using CCC codes, which captures diagnostic codes retrospectively and throughout a hospitalization, using our screening algorithm, we categorized a child’s medical complexity at the time of admission. We suspect that some children we classified as PH or CSHCN at the time of admission may have progressed in medical complexity during the remainder of the hospitalization; in this situation, their medical complexity level may have been lower in our approach, compared to a retrospective approach. For example, 15% of children classified as PH by our screening algorithm had at least 1 CCC code, likely because of newly diagnosed or acquired conditions during a subsequent hospitalization. Despite these differences, we found similar resource use for children identified as CMC by using our screening and the diagnostic code definition of ≥4 CCC codes.

In previous studies, researchers have shown hospital use to increase with greater number of CCCs18 ; similarly, we found our population of CMC had comparable inpatient resource use to children with ≥4 CCC qualifying codes. In contrast to the literature, we did not find resource use to be significantly different when adding measures of NI or TA to a CCC code definition (≥4) of CMC.1  We suspect this may be due to limitations of our data, which did not include outpatient use or inpatient use at other institutions. More comprehensive payer claims data that include outpatient and outside facility use may detect differences in use in this subpopulation of children with ≥4 CCC codes. Alternatively, children with NI and/or requiring TA often have chronic illnesses or disabilities that originated in the neonatal period; thus, they are likely to have already been established with outpatient resources that could, in fact, have limited their need for hospitalization.

In contrast to the literature, we found relatively high concordance of TA between diagnostic and procedure code–based identification of TA and our point-of-care chart review screening.19  For example, when comparing TA by using screening versus diagnostic codes, we found 100% vs 97% of CMC, 21% vs 29% of CSHCN, and 0% vs 3% of PH children required TA. Of note, CSHCN and PH children were more likely to be identified as using technology through using diagnostic codes, compared to our definition. This likely reflects 2 factors: (1) inclusion of more types of technology in the diagnostic code definition (ie, evacuation tubes, renal support, and cardiac support) than the shorter list of technology devices used in our point-of-care screening criteria and (2) assignment of TA status at admission by using our point-of-care screening will inherently omit children with new technology devices placed later in a given admission.

Lastly, we found our point-of-care complexity screening algorithm for CMC, embedded in a complex care program, to be feasible. The algorithm described was performed with daily chart review by a single program coordinator, taking ∼90 minutes per day for an average of 10.7 patients per day, with few discordant screenings. Although our screening tool was used to identify CMC during an inpatient hospitalization, our approach to identify CMC could also be applicable and feasible in an outpatient setting. In the outpatient setting, screening could be applied to a clinic schedule in advance because it is performed exclusively by using chart review. Alternatively, screening could be performed at the time of a clinical encounter. Additionally, our screener could be used in combination with CCC or PMCA approaches. For example, if we had used >1 CCC as a prescreener embedded in the electronic health record, our program coordinator would only have had to screen 940 instead of 2187 charts and would have picked up 76 of the 77 CMC. In addition, studies are needed to determine the feasibility of this approach in the outpatient context and validate the generalizability of this screening tool to other institutions.

This study had several limitations. First, the algorithm was applied at a single pediatric tertiary care center within5 an existing pediatric complex program. These features limit generalizability to other institutions without complex care programs or with different complex care program structures because these programs are increasingly common nationwide.20  Second, our inpatient-based screening algorithm misses CMC that might benefit from care coordination (to prevent hospitalization and complications) but have not had any recent acute hospitalizations in the past 12 months. CMC without recent hospitalizations because of effective outpatient care management within a primary care-based care delivery model may have different needs and require a different approach to identification than hospitalized CMC.21  In this study, we used high hospital use as a proxy for high intensity care coordination needs. However, the goals of care coordination are more expansive than reduction of hospital use; use measures alone do not capture the positive impact of care coordination on the financial, emotional, and social needs of CMC patients and families.22  Thus, there are CMC who have not yet incurred hospital costs to date but are at risk for future hospitalizations and can benefit now from the multiple positive impacts of enhanced care coordination. Expected clinical benefit despite lack of high hospital use to date was the main reason why 4 of 77 CMC (5.2%) did not meet the high hospital use criterion yet were enrolled into the complex care program after further case reviews. Third, our data on hospital use represented inpatient use only at our institution. It did not include outpatient, urgent care, subspecialty, or emergency department costs nor did it reflect inpatient costs at other nearby institutions. Lastly, the length of time a child was managed in our study was not standardized and depended on when in FY16 a child was screened. This could impact our data if one group of children (ie, CMC) were screened earlier in FY16 and had more time to accrue use. In future studies, researchers could use payer claims to provide a more comprehensive overview of health care spending.

With this study, we add to the growing body of literature focused on strategies to identify CMC. To our knowledge, this was the first study of a point-of-care screening algorithm to identify CMC with high health care use. This algorithm can be used to target a population of CMC that would benefit most from the care coordination services provided by a dedicated complex care program.

We thank John Owens, RN, Stephanie Smith, RN, BSN; Stephanie Whitfield, RN, BSN; Natalie Krohl, RN, PNP; Mikelle Key-Solle, MD; Heather S. McLean, MD; and Kathleen W. Bartlett, MD, for their contributions.

Dr Parente conceptualized and designed the study and drafted the initial manuscript; Drs Ming, Parnell, and Childers assisted with study design and reviewed and revised the manuscript; Ms Spears performed all statistical analyses; Ms Jarett performed data collection and reviewed the manuscript; and all authors approved the final manuscript as submitted.

FUNDING: Dr Ming’s contributions were supported in part by the National Institutes of Health National Heart, Lung, and Blood Institute grant K12HL13830. Funded by the National Institutes of Health (NIH).

1
Cohen
E
,
Berry
JG
,
Camacho
X
,
Anderson
G
,
Wodchis
W
,
Guttmann
A
.
Patterns and costs of health care use of children with medical complexity
.
Pediatrics
.
2012
;
130
(
6
). Available at: www.pediatrics.org/cgi/content/full/130/6/e1463
2
Berry
JG
,
Hall
M
,
Neff
J
, et al
.
Children with medical complexity and Medicaid: spending and cost savings
. [
published correction appears in Health Aff (Millwood) 2015;34(1):189
].
Health Aff (Millwood)
.
2014
;
33
(
12
):
2199
2206
3
Kuo
DZ
,
Goudie
A
,
Cohen
E
, et al
.
Inequities in health care needs for children with medical complexity
.
Health Aff (Millwood)
.
2014
;
33
(
12
):
2190
2198
4
Cohen
E
,
Kuo
DZ
,
Agrawal
R
, et al
.
Children with medical complexity: an emerging population for clinical and research initiatives
.
Pediatrics
.
2011
;
127
(
3
):
529
538
5
Casey
PH
,
Lyle
RE
,
Bird
TM
, et al
.
Effect of hospital-based comprehensive care clinic on health costs for Medicaid-insured medically complex children
.
Arch Pediatr Adolesc Med
.
2011
;
165
(
5
):
392
398
6
Gordon
JB
,
Colby
HH
,
Bartelt
T
,
Jablonski
D
,
Krauthoefer
ML
,
Havens
P
.
A tertiary care-primary care partnership model for medically complex and fragile children and youth with special health care needs
.
Arch Pediatr Adolesc Med
.
2007
;
161
(
10
):
937
944
7
Mosquera
RA
,
Avritscher
EBC
,
Samuels
CL
, et al
.
Effect of an enhanced medical home on serious illness and cost of care among high-risk children with chronic illness: a randomized clinical trial
.
JAMA
.
2014
;
312
(
24
):
2640
2648
8
Berry
JG
,
Hall
M
,
Cohen
E
,
O'Neill
M
,
Feudtner
C
.
Ways to identify children with medical complexity and the importance of why
.
J Pediatr
.
2015
;
167
(
2
):
229
237
9
Simon
TD
,
Cawthon
ML
,
Popalisky
J
,
Mangione-Smith
R
;
Center of Excellence on Quality of Care Measures for Children with Complex Needs (COE4CCN)
.
Development and validation of the Pediatric Medical Complexity Algorithm (PMCA) version 2.0
.
Hosp Pediatr
.
2017
;
7
(
7
):
373
377
10
Simon
TD
,
Cawthon
ML
,
Stanford
S
, et al
.
Pediatric Medical Complexity Algorithm: a new method to stratify children by medical complexity
.
Pediatrics
.
2014
;
133
(
6
). Available at: www.pediatrics.org/cgi/content/full/133/6/e1647
11
Simon
TD
,
Haaland
W
,
Hawley
K
,
Lambka
K
,
Mangione-Smith
R
.
Development and validation of the Pediatric Medical Complexity Algorithm (PMCA) version 3.0
.
Acad Pediatr
.
2018
;
18
(
5
):
577
580
12
Ming
DY
,
Jackson
GL
,
Sperling
J
, et al
.
Mobile complex care plans to enhance parental engagement for children with medical complexity
.
Clin Pediatr (Phila)
.
2019
;
58
(
1
):
34
41
13
Harris
PA
,
Taylor
R
,
Thielke
R
,
Payne
J
,
Gonzalez
N
,
Conde
JG
.
Research Electronic Data Capture (REDCap)–a metadata-driven methodology and workflow process for providing translational research informatics support
.
J Biomed Inform
.
2009
;
42
(
2
):
377
381
14
van Dyck
PC
,
McPherson
M
,
Strickland
BB
, et al
.
The National Survey of Children with Special Health Care Needs
.
Ambul Pediatr
.
2002
;
2
(
1
):
29
37
15
Feudtner
C
,
Feinstein
JA
,
Zhong
W
,
Hall
M
,
Dai
D
.
Pediatric complex chronic conditions classification system version 2: updated for ICD-10 and complex medical technology dependence and transplantation
.
BMC Pediatr
.
2014
;
14
:
199
16
Peltz
A
,
Hall
M
,
Rubin
DM
, et al
.
Hospital utilization among children with the highest annual inpatient cost
.
Pediatrics
.
2016
;
137
(
2
):
e20151829
17
Simon
TD
,
Berry
J
,
Feudtner
C
, et al
.
Children with complex chronic conditions in inpatient hospital settings in the United States
.
Pediatrics
.
2010
;
126
(
4
):
647
655
18
Berry
JG
,
Hall
DE
,
Kuo
DZ
, et al
.
Hospital utilization and characteristics of patients experiencing recurrent readmissions within children's hospitals
.
JAMA
.
2011
;
305
(
7
):
682
690
19
Berry
JG
,
Goodman
DM
,
Coller
RJ
, et al
.
Association of home respiratory equipment and supply use with health care resource utilization in children
.
J Pediatr
.
2019
;
207
:
169.e2
175.e2
20
Cohen
E
,
Berry
JG
,
Sanders
L
,
Schor
EL
,
Wise
PH
.
Status complexicus? The emergence of pediatric complex care
.
Pediatrics
.
2018
;
141
(
suppl 3
):
S202
S211
21
Pordes
E
,
Gordon
J
,
Sanders
LM
,
Cohen
E
.
Models of care delivery for children with medical complexity
.
Pediatrics
.
2018
;
141
(
suppl 3
):
S212
S223
22
Kuo
DZ
,
McAllister
JW
,
Rossignol
L
,
Turchi
RM
,
Stille
CJ
.
Care coordination for children with medical complexity: whose care is it, anyway?
Pediatrics
.
2018
;
141
(
suppl 3
):
S224
S232

Competing Interests

POTENTIAL CONFLICT OF INTEREST: The authors have indicated they have no potential conflicts of interest to disclose.

FINANCIAL DISCLOSURE: No honorarium, grant, or other form of payment was given to any of the authors to perform this work or produce this article. The authors have indicated they have no financial relationships relevant to this article to disclose.

Supplementary data