Factors Correlated With Successful Pediatric Post-Discharge Phone Call Attempt and Connection Free

Patients aged 0 to 18 years, who were discharged to home from pediatric hospitalist services at 4 Project IMPACT pilot sites between January 2014 and December 2017, were included for analysis in this study. Patients discharged from the newborn nursery or pediatric, neonatal, or cardiac intensive care units were not included. A postdischarge call placed within 3 days of hospital discharge to the discharged patient’s primary caregiver was 1 of 4 Project IMPACT hospital-to-home bundle elements.³  Each pilot site had a local improvement team and used multiple planned sequential interventions to implement the bundle elements. The purpose of the postdischarge call was to determine discharge readiness, identify discharge errors, questions, and misunderstandings, and reinforce medication and follow-up plans. Three call attempts were standard for each discharged patient. Although the postdischarge call questionnaire was uniform across sites, further determination of the post-discharge call model in terms of who performed the call, financial support for the caller, and bandwidth for procedural elements differed according to local context and varied between sites (Fig 1).

Chart review of demographic information and record of postdischarge call documentation was performed for each patient and entered in a Research Electronic Data Capture (REDCap) database.¹⁴  Data were entered into the REDCap database by study investigators and monitored for completion and accuracy by site leaders (including D.C., S.G., L.M., and S.N.O.). Demographic variables documented were patient sex, age, caregiver’s preferred language, payor type, and hospital site. Payor types included federal or state payor, private insurance, or self-pay or uninsured. Payor types that did not fall into 1 of these 3 categories were noted as “other payor type.” Clinical variables documented were clinical risk group (CRG), patient use of technology, readmission within 3 days, and hospital length of stay. CRG classification fell into 3 categories based on previously published definitions: no chronic disease (CRG1; eg, patient with bronchiolitis), single chronic condition (CRG2; eg, patient with asthma), and multiple or complex chronic conditions, including technology-supported patients (CRG3).¹⁵  CRG group determination was performed by the investigator reviewing the electronic health record (EHR) based on the patient’s problem list, medical history, and technology use. Investigators received standardized training on CRG classification, new trainee classifications were audited for accuracy by site leaders, and equivocal cases were resolved by consensus. We defined use of technology as any patient discharged with a new or preexisting gastrostomy tube, ventriculoperitoneal shunt, central venous line (including peripherally inserted central catheters), and/or tracheostomy. Procedural variables documented were phone call attempt and successful connection, which were binary, categorical (yes or no) variables. “Phone call attempt” was defined as any attempt by a caller to reach the caregiver of a discharged patient via postdischarge phone call, while “successful connection” was defined as a caller reaching and speaking with a caregiver of a discharged patient via postdischarge phone call.

We summarized categorical data as proportion (n, %) and continuous data as mean (SD) for normally distributed data or median (range) for nonnormally distributed data. Age data were stratified into 4 groups (≤30 days, 31 days–1 year, >1 year–6 years, >6 years) and analyzed as a categorical variable. Age categories used reflect common pediatric clinical groups, including neonates, infants, patients in early childhood, and school-aged children. School-aged children included ages 6 through 18, as we hypothesized that differences in call connectivity were driven more by caregiver characteristics than patient characteristics for these ages. We compared demographic, clinical, and procedural variables between subgroups (call attempt made or not made; successful connection or nonsuccess among those with a call attempt) using Mann Whitney U test for continuous data and either χ² test with or without continuity correction or Fisher’s Exact test for categorical data.

We evaluated homogeneity among hospital sites caused by known differences in their connectivity models and populations served. Homogeneity among hospital sites was evaluated by the Breslow-Day test for categorical data and by univariate analysis of variance (ANOVA) for continuous data.

To identify significant independent predictors of both call attempt and successful connection, all variables that achieved a relationship (P < .1) with each factor in univariate analyses were entered as independent variables into a mixed model logistic regression model, as appropriate, with either call attempt made (yes or no) or successful connection (yes or no) as the dependent variable and hospital site entered as random effect. For categorical covariates, the reference groups were age >6 years, female sex, English as caregiver’s preferred language, absence of technology at discharge, nonchronic disease (CRG1), no readmission within 3 days, and payor other than federal or state. Unadjusted odds ratios were calculated by evaluating each covariate in a separate mixed model logistic regression. Collinearity was evaluated by linear regression after creating dummy variables to express all subgroups of categorical data binary variables; a variance inflation factor >3 was used to define multicollinearity among covariates. Statistical analyses were performed using SPSS Statistical Software, version 28 (IBM SPSS Inc, Armonk, NY).

Central institutional review board approval was obtained from the primary investigator’s institution. Site-specific institutional review board approval or exemption was obtained.

Postdischarge calls were attempted for 5528 of 7725 (71.6%) total patient discharges across the 4 hospital sites (Fig 1). Attempt rates were highest for Site A (88.8%), followed by Sites B and C (78.4% and 74.1%, respectively), and lowest for Site D (43.6%) (Table 1). Supplemental Table 3 shows site-specific data for the considered demographic, clinical, and procedural variables. There was significant heterogeneity among sites for the relationship between call attempt and caregiver language (P = .003), patient age (P < .001), technology use (P < .001), and readmission within 3 days (P = .029).

In unadjusted analysis, the relationships between all considered demographic, clinical, and procedural variables (except sex and length of stay) and phone call attempt varied significantly (Table 1).

Following adjusted analysis, age less than 30 days (P = .03), age 1 to 6 years (P = .04), and English language (P < .001) showed significant positive independent associations with call attempt (Table 2). Readmission within 3 days was inversely related to call attempt (P = .004) (Table 2).

Overall, there was successful connection for 3801 of 5528 (68.8%) calls attempted (Fig 1). Connection rates varied significantly among the 4 sites (52% to 79%, P < .001) (Table 1). Supplemental Table 4 shows site-specific data for the considered demographic, clinical, and procedural variables. There was significant heterogeneity among sites for the relationship between completed calls and both caregiver language (P = .002) and patient age (P < .001).

In unadjusted analysis, the relationships between all considered demographic, clinical, and procedural variables (except English language, readmission, and length of stay) and successful call connection varied significantly (Table 1).

Following adjusted analysis, age less than 30 days (P = .01), age >1 to 6 years (P = .04), and chronic noncomplex CRG2 (P = .02) were significant positive predictors of successful connection, whereas federal or state payor type (P = .02) was inversely related with successful connection Table 2).

Our analysis of postdischarge calls for over 7700 patient discharges across 4 hospitals using a variety of call models shows that there are significant differences according to demographic and clinical patient characteristics for both postdischarge calls attempted and for successful connection for those with attempted calls. After adjusted analysis, calls were more frequently attempted in the children aged 0 to 6 years, to caregivers who prefer English, and to those not readmitted within 3 days of discharge. After adjusted analysis, successful call connection was significantly associated with age 0 to 6 years, the chronic noncomplex clinical risk group (CRG2) and nonfederal or state payors. These findings are important, as postdischarge calls are frequently used as an intervention to prevent hospital-to-home transition failures. Lack of call attempt and successful connection may represent missed opportunities to improve patient safety.

In a qualitative study exploring the transitional needs of pediatric caregivers, Solan et al found that ensuring proper understanding of discharge instructions, return precautions, medication administration, and follow-up plans is important, and that caregivers report limited ability to process all necessary information at the time of discharge.¹⁶  It may be particularly important to discuss discharge information with caregivers of vulnerable populations, like those with complex medication regimens or technology support. Although they have not been shown to improve readmission rates, effect of postdischarge calls has been explored as an intervention by examining outcomes, such as number of corrected misunderstandings or changes in patient satisfaction scores.^8,11,13  In a study with one of the highest published connection rates (78%), Heath et al found that 20% of caregivers needed an issue addressed during the postdischarge call, reinforcing that a postdischarge call may improve patient safety during hospital-to-home transitions.¹³  However, studies have yet to compare the demographic, clinical, or procedural data of the populations that are associated with call attempt by the care team and successful connection when an attempt is made. We suggest understanding the differences in these populations may inform prospective efforts to increase postdischarge call connectivity and lead to consideration of connection rates when interpreting hospital-to-home outcome data obtained from calls.

In our study, call attempt and connectivity rates differed widely among the 4 study sites (44% to 89% and 52% to 79%, respectively). Our connection rates are generally consistent with prior studies.^9,13,11,17  Although each site aimed to standardize the number and timing of call attempts and the questions used during calls, local context defined call models and capacity at each of our 4 study sites, contributing to variability in call attempt and connection rates between sites. Of the 4 Project IMPACT bundle elements, the postdischarge call presents the largest additional resource burden.³  Sites A, B, and C had strategies that included an aim to reach all patient discharges, whereas site D targeted patients with medical complexity. Sites with the highest connection rates (sites A and D) both had a single designated nurse making most of the calls (Fig 1). Site B had the lowest connection rate, likely because of loss of funding for their designated caller part way through the study (Supplemental Fig 2). This finding may support that investment in a reliable and consistent program, including a dedicated caller, can optimize connection rates.

Following adjusted analysis, caregivers who prefer English were more likely to have a postdischarge call attempted. This represents a potential healthcare disparity that may be attributed to the burden of additional time and resources required to use appropriate interpreter services, which are known to improve the quality of clinical care patients with limited English proficiency receive.¹⁸  Patients preferring non-English languages are known to be at increased risk for patient safety events¹⁹  and may benefit from the postdischarge call’s ability to identify and correct misunderstandings.

In adjusted analysis, successful call connection was not significantly different between caregivers who preferred English and non-English languages. However, there was variability in connection rates by site (Supplemental Table 4). Interestingly, Site D’s trend differed from other sites with higher connectivity for those that preferred non-English languages (80.4%) than for those that prefer English (65.9%) (Supplemental Table 4). Key differences in the call model for this site included prioritization of non-English speaking families by the caller and role of the caller as an embedded care coordinator in the outpatient complex care medical home. It is possible that caregivers are more likely to answer the phone when called by a familiar and trusted member of their child’s primary care team. As institutions seek to increase equity in care provided across populations, our findings may support investment in targeted approaches to address gaps in care related to language differences, thus resulting in improved patient outcomes for families that do not prefer English.

In addition to language preference other than English, populations with other social needs, such as lack of secure housing or phone services, may be more difficult to connect with and may also be at higher risk for morbidity.^20,21  Our adjusted analysis showed that although likelihood of call attempt did not differ, federal and state payor type was inversely associated with successful call connection. Project IMPACT did not collect data on household income, so this was the closest socioeconomic surrogate in our study. Future studies may consider use of zip codes or geocoding as a better correlate.²²  Enhanced predischarge efforts to improve likelihood of contact, such as social needs screening and mitigation, prospective agreement on contact information, identifying a time for the call either at the time of discharge or through text messaging,²³  and/or giving the families a dedicated number to call to initiate contact with available voicemail services to facilitate messages may be indicated.

Our adjusted postdischarge call analysis shows that children with chronic noncomplex medical conditions (CRG2) are more likely to have successful connection and children with chronic complex medical conditions (CRG3) trend toward increased likelihood of successful connection. Currently, there is no available literature to shed light on this finding. It may be that caregivers of chronically ill children are used to keeping in close contact with healthcare providers, are more likely to value the opportunity to clarify and ask questions, or are more likely to perceive the child as not well or at higher risk for morbidity as compared with caregivers of children without chronic disease. More research is required to understand this association.

The inverse relationship between call attempt and readmission within 3 days after adjusted analysis is potentially because the child had returned to the hospital, an event that would be apparent in the EHR and would eliminate the need for a postdischarge call within the same timeframe. Successful connection has not been found to be associated with decreased readmission.¹¹  However, readmission rate is likely not the outcome by which effect of postdischarge calls should be measured, as they are primarily a safety tool.

Limitations to our study included our heterogeneous sample and variability of call models across the sites, which were constrained by resources and practicality of implementation as part of a quality improvement initiative. However, the sample and call model variability across sites reflects reality and supports the application of our data in a diverse array of postdischarge call programs. A limitation of our data collection method is the potential for human error during original EHR documentation, chart review and entry into REDCap, or CRG classification. This is demonstrated in the lack of consistency in documenting number of call attempts. Thus, we cannot comment on if more call attempts versus any call attempts are associated with successful connection. Finally, as our available data are from 2014 to 2017, our analysis reflects precoronavirus pandemic communication trends. It is possible that postdischarge call attempt and connection associations may be different postpandemic, as technology preferences and trends change. However, there will likely always be a role for postdischarge calls, as access to or competence in newer communication strategies may not be universal.

Further studies are needed to assess for healthcare disparities related to postdischarge call programs and interventions to improve pediatric postdischarge call connection rates. Additionally, analysis of the value of postdischarge calls could help programs advocate for resources to implement them as part of hospital-to-home transition care or to select which patients are most likely to benefit from this investment.

In our retrospective multicenter analysis, we identified that across 4 hospital sites postdischarge calls were attempted more often for caregivers with children aged 6 or younger and those who preferred English and less often for patients readmitted within 3 days. For patients with calls attempted, successful connection was less likely for school aged children and federal or state payor type and more likely for children with noncomplex chronic disease (CRG 2). Our analysis provides potential directions for prospective interventions to improve call attempt and successful connection rates. Call attempt rates may be increased by investment in a reliable call program model or improvement in interpreter services to address potential healthcare disparities. Successful connection may be increased by considering the caller’s role in the medical team or integrating population-based communication strategies based on age, medical complexity, or social needs of the patient.

We thank the IMPACT Study Group; the Department of Pediatrics, Tufts University School of Medicine and The Barbara Bush Children’s Hospital; The Medical College of Wisconsin and Children’s Hospital of Wisconsin; the Department of Pediatrics, Weill Cornell Medicine and New York-Presbyterian Komansky Children’s Hospital; and the Department of Pediatrics, Drexel University College of Medicine and St Christopher’s Hospital for Children.