Medications Reconciled at Discharge Versus Admission Among Inpatients at a Children’s Hospital

The study was conducted at The Hospital for Sick Children (SickKids), a freestanding 350-bed tertiary care children’s hospital in Toronto, Ontario, Canada. Data were collected for admissions from January 1, 2016, to December 31, 2017, including data through to January 10, 2018. During that period, SickKids patients’ medications were routinely reviewed and reconciled on admission to the hospital by nurses and/or physicians and verified by pharmacists. At discharge, home medications were reconciled, and a list of home medications was generated.

Data were extracted from electronic medical records and a discharge diagnosis database. Inpatient encounters were included if they met the following criteria: age ≤17 years on the day of admission; admitted from home or via the emergency department (ED); and discharged from the hospital. Admissions transferred from other hospitals’ inpatient units were excluded. Admissions including time in the NICU were excluded because most such infants had not been home since birth. Admissions were further excluded if medication reconciliation data were missing or left incomplete by the health care provider who last accessed the document. Patients could contribute multiple admissions to the analyses.

Medications were classified as admission medications if they were listed in the admission reconciliation form and were not labeled as new prescriptions initiated by the admitting service. “Admission medications” included home medications and medications started in the hospital but before admission, such as in the ED. Discharge medications included those prescribed in the hospital that were intended to be continued at home and medications prescribed newly at discharge. Each discrete medication entry was coded by the generic drug name and classified on the basis of the American Hospital Formulary System (https://www.ahfsdruginformation.com/ahfs-pharmacologic-therapeutic-classification/). Each medication was counted once in a single reconciliation even if it was listed as multiple entries (eg, same drug prescribed at 2 different doses).

Patients were categorized by sex and age on the date of admission (infancy [≤1 year of age], early childhood [1–6 years of age], and late childhood and adolescence [6–17 years of age]). Presence of a complex chronic condition (CCC) was based on International Classification of Diseases, 10th Revision, diagnostic codes at discharge, per Feudtner et al.^21,22  Each admission was further classified by discharge clinical service: general medicine, subspecialty medicine, psychiatry and adolescent medicine, surgery, and critical care (Supplemental Fig 6). Other covariates included admission to the PICU at any time during the hospitalization and admissions with at least 1 surgical procedure during the hospitalization.

Distributions of medication counts at admission and discharge were summarized by using medians, interquartile ranges (IQRs) and minimum and maximum, and the percentage of patients at admission and discharge with medication counts of 0, ≥2, and ≥5 medications, overall and in subgroups defined by patient- and hospitalization-related covariates. We identified the most common drug or medication subclasses at discharge and calculated the proportions of discharges at which specific drugs or subclasses were listed, further disaggregated according to whether they were new or continuing (as defined above). Changes in medication number from admission to discharge (overall and within patient subgroups) were estimated by multilevel negative-binomial regression with robust SEs and random intercepts to account for clustering of paired admission and discharge events within a hospitalization and multiple hospitalizations of the same patient. The exponentiated regression coefficient for a fixed-effect “time point” indicator variable (admission or discharge) was interpreted as the relative change (RC) in medication count, with 95% confidence intervals (CIs), from admission to discharge. To examine the relative risk (RR) that a specific drug or drug subclass was listed at discharge (versus admission), we used Poisson multilevel models with robust SEs, with random intercepts to account for clustering, as for the negative-binomial models.²³  For both the RC and RR models, we omitted the patient-level random intercept if the model failed to converge with 2 levels.

We interpreted RCs in medication count (negative-binomial models) or RRs of a drug or subclass at discharge (Poisson models) as the effects of inpatient admission and any related changes in health status or diagnoses arising during that admission, assuming negligible confounding by other patient characteristics that do not change substantially over the course of most admissions. Models were extended to evaluate the modifying effects of covariates on the RC or RR by including covariate-time point interaction terms, whereby P < .05 was considered to indicate a statistically significant interaction. Effects of covariates (CCC status, surgery during hospitalization, PICU admission during hospitalization, sex, age, and discharge service) were not assumed to be causal.

RC estimates were first generated for an “unrestricted” analysis. Subsequent analyses excluded certain medications in a stepwise manner (referred to as “levels of restriction”) to remove medications that are mainly dispensed over-the-counter and those that are mainly used for fixed short durations, to focus on medications that were more likely to contribute to polypharmacy and/or adverse events. Level 1 restriction excluded vitamins, minerals, and natural or homeopathic products; level 2 restriction excluded over-the-counter analgesics, such as acetaminophen, ibuprofen, and acetylsalicylic acid, in addition to level 1 medications; and level 3 restriction removed cathartics and laxatives (eg, polyethylene glycol and lactulose) in addition to level 1 and 2 medications. Level 4 restriction was the primary analysis for which we removed antibiotics, in addition to level 1, 2, and 3 medications. In sensitivity analyses, we examined the impact of missing data on selected models using multiple imputation by chained equations (MICEs) and repeated selected analyses, including hospitalizations with incomplete and missing admission medication reconciliation documents. Analyses were performed by using Stata 16 (Stata Corp, College Station, TX).

A subset of medical records (n = 200) underwent manual review to verify the presence of medication reconciliation documents, the number of medications, and the listing of morphine at discharge. An initial review (n = 100 records) led us to identify and resolve a coding issue related to the classification of a document as missing; in a second round of reviews (n = 100), this discrepancy was confirmed to have been resolved. In both review rounds, we did not find any discrepancies between the electronic database and manual review related to the numbers of medications or whether morphine was listed.

Among 32301 encounters screened for inclusion (15 729 in 2016; 16 572 in 2017), 10 243 encounters (32%) were ineligible for analyses, yielding a final data set with 22058 hospitalizations of 15 268 unique patients (Supplemental Fig 7). Among hospitalizations that were otherwise eligible, 29% (6093 of 28152) were excluded from primary analyses because of missing or incomplete reconciliation documentation. Medication counts varied considerably across subgroups defined by patient and hospitalization characteristics, with an overall median of 2 (IQR: 0–4) medications at admission and 3 (IQR: 1–5) at discharge, and 75% had at least 2 medications at discharge (Table 1).

In the primary analysis (level 4 restriction excluding nonprescription medications and antibiotics), there was a significant effect of hospitalization on medication counts at discharge versus admission (RC: 1.27 [95% CI: 1.25–1.29]). The change in medication count from admission to discharge was observed at all other levels of restriction and in all strata (Fig 1). With the exception of patient sex, all other covariates modified the magnitude of the change in medications at discharge versus admission (Fig 1): infants (compared with both older age groups) had higher RC estimates across all levels of restriction; the presence of a CCC decreased the RC (although the interaction term was nonsignificant at the fourth level of restriction); having at least 1 surgical procedure increased the RC across all levels of restriction; admission to the PICU during hospitalization increased the RC across all restriction levels; and discharge from a surgical service increased the RC, compared with general medicine across all restriction levels (Supplemental Table 2).

In the primary analysis, the most common subclasses at discharge were opioids, proton pump inhibitors (PPIs), bronchodilators, antiemetics, and corticosteroids (Fig 2), and the most common specific medications at discharge were morphine, salbutamol, omeprazole, lansoprazole, and ondansetron (Fig 3). Overall, opioids were listed at 5.4% and 22% of admissions and discharges, respectively (RR: 4.0 [95% CI: 3.73–4.32]; Fig 4). A similar proportion of opioids were listed at admission and discharge when including discharges with incomplete or missing admission medication reconciliation documents (Supplemental Table 5). The relative increase in opioid prescriptions at discharge was similar across sex and age groups. Hospitalizations with at least 1 CCC were less likely to list opioids at discharge (15%) and had a lower RR at discharge versus admission (1.89 [95% CI: 1.75–2.04]), compared with those without a CCC (21%; RR: 6.89; [95% CI: 6.28–7.57]; Fig 4). Surgery during hospitalization increased the probability of discharge with an opioid prescription (68%; RR: 13.3 [95% CI: 11.5–15.3)], compared with that of those who did not have surgery (13%; RR: 2.38 [95% CI: 2.22–2.56]; Fig 5). Discharge from a surgical service increased the RR of discharge with an opioid prescription (Fig 4); of the patients discharged on opioids from a surgical service, 39% had not had surgery during the hospitalization. Morphine was the most common opioid prescribed at discharge (RR: 4.54 [95% CI: 4.18–4.92] versus admission); like the opioid subclass, the RR for morphine was modified by a CCC, surgical procedure, and discharge from a surgical service (Fig 5). The most common primary diagnoses on discharge with an opioid prescription were supracondylar fracture of the humerus, sickle cell crisis, tonsillectomy, cleft palate repair, and cholesteatoma of the middle ear (data are not shown).

PPIs were more likely to be listed at discharge versus admission (18% versus 15%, respectively; RR: 1.21 [95% CI: 1.19–1.24]). The RR of a PPI prescribed at discharge was lower in those who had at least 1 surgery and increased in infants, those who were in the PICU. and those discharged from a subspecialty medical service (Fig 4). Omeprazole prescriptions at discharge increased to a lesser extent in infants, whereas lansoprazole prescriptions increased to a greater extent in infants (Fig 5). The RRs for both PPIs were lower in patients who underwent surgery and higher in patients who were in the PICU, and the RR for omeprazole was increased in those with a CCC (Fig 5). The effects of hospitalization on bronchodilator, antiemetic, and corticosteroid prescriptions were variable across subgroups (Figs 4 and 5).

Sensitivity analyses using MICE models for selected analyses (overall opioid use, opioid use stratified by surgery, and level 4 medication counts) supported inferences from the primary complete case analysis (Supplemental Table 6).

At a large Canadian children’s hospital, the number and type of drugs prescribed at discharge differed substantially from those reported at the onset of inpatient admission, indicating the important contribution of hospitalization to home medication regimens. Overall and across all patient strata, the number of medications at discharge was higher than at admission, even after excluding antibiotics and most over-the-counter drugs. Specific drugs and subclasses that were most frequently prescribed at discharge in the current study were similar to those in which children were commonly exposed to during inpatient admission in a study of children’s hospitals in the United States.⁹  Although new diagnoses during hospitalization were expected to lead to initiation of therapies continued at home, the most commonly prescribed medications at discharge were those typically used to address symptoms rather than underlying diseases (opioids, PPIs, and antiemetics). Although indications or appropriateness of prescriptions were not assessed in the study, the frequent use of such medications may reflect nonessential or discretionary prescribing (ie, the propensity to prescribe varies across physicians, suggesting the medication may be unnecessary in some cases). These findings indicate discharge prescribing is a potential target of interventions to promote rational prescribing and deprescribing in pediatrics to reduce medication overuse and polypharmacy in the outpatient setting.^1,2 

Opioid prescriptions were much more likely at discharge, compared with admission, particularly in postsurgical patients. Most patients (68%) who underwent surgery were discharged on opioids, as were about one-half (47%) of all patients discharged from a surgical service, including many patients who had not had surgery during the hospitalization. Opioid prescriptions are of particular concern because of their association with persistent²⁴  and future²⁵  opioid misuse in adolescents. There has been a marked rise in pediatric opioid prescriptions in the United States over the past 2 decades²⁶  and an increase in pediatric mortality because of opioid toxicity.²⁷  Particular attention has been given to opioid overuse in the postsurgical setting.^28,29  In the inpatient setting, longer lengths of stay, surgery, and PICU admission were previously found to be risk factors for opioid exposure.³⁰  With the present findings, we support the need to integrate opioid stewardship with pediatric surgical discharge planning to promote selection of nonopioid alternatives as discharge analgesics³¹  and highlight the importance of coordination with primary care physicians and outpatient follow-up. This is consistent with current SickKids Choosing Wisely recommendations, which included guidance to treat pain with multimodal therapeutic approaches, rather than opioids, when possible.³² 

PPIs were frequent discharge medications, particularly among infants, patients with CCCs, and patients admitted to PICU; however, many patients were already taking PPIs at admission. Overuse of PPIs in infancy has been widely recognized³³  because PPIs are commonly prescribed to treat symptoms of gastroesophageal reflux, despite the risk of harm and lack of evidence of therapeutic benefit in this age group.^34,35  Hospitalization may represent an opportunity to facilitate deprescribing of PPIs and promote the avoidance of new discretionary PPI prescriptions, particularly in infants.

Although children with CCC had higher prevalence of medication use at admission, hospitalization had less of an effect on discharge (versus admission) medications among children with CCCs, compared with those without a CCC. We speculate that the participation of many such children in a structured outpatient complex care programs in the SickKids catchment area^36,37  may have reduced the extent to which changes to regular medication regimens occurred during inpatient admissions but acknowledge that such programs may not be widely available in other regions. Conversely, children admitted to the PICU experienced greater average increases in the overall number medications prescribed at discharge and were especially more likely to be prescribed PPIs and corticosteroids at discharge. Dai et al⁷  found that PICU patients were exposed to an average of 20 medications over the course of their hospitalization. Some medications initiated in the hospital may be appropriately continued at discharge (eg, weaning corticosteroids), yet, with the results, we highlight the importance of early initiation of weaning protocols and other opportunities to deprescribe in the critical care environment.

There were several limitations to this study. The data set covered a recent 2-year period at a large pediatric hospital likely to be representative of many other large academic pediatric centers in North America but may have limited application to pediatric patients treated in general hospitals.⁹  The medication reconciliation process uses multiple sources, including parental recall, community pharmacies, and previous hospitalizations and was ideally verified by at least 2 sources during reconciliation, but errors in the reconciliation process could have influenced our findings.^17,19  A strength of the study was the availability of data from paired admission and discharge medication reconciliation documents, but data were often missing or incomplete; however, our inferences did not change when we used multiple imputation to address missingness. Because we were unable to distinguish “home” medications at admission from those prescribed in or just before the ED, the magnitude of the increase at discharge was likely a conservative estimate of the true effect of hospitalization on home medications. We could not determine if the discharge prescriptions were filled or quantify the intended or actual duration of discharge medication use; therefore, we could not distinguish the medications that were intended to be continued long-term and, therefore, most likely to contribute to polypharmacy. It was beyond the scope of the study to adjudicate the appropriateness of prescriptions, examine between-prescriber variations, or assess adverse risks of medications or polypharmacy (eg, drug–drug interactions). For some conditions, there may be a trade-off between length of stay and the number of medications at discharge; for example, proactive discharge before symptom resolution or completion of therapy may reduce in-hospital resource use^38,39  but could also increase the number of home medications (eg, oral stepdown therapy and prescriptions for symptom management). Although we were unable to address this trade-off in the current study, future research should examine the broad effects of variations in discharge prescribing practices, including potential cost savings and psychosocial benefits of interventions that reduce the length of inpatient stay by enabling the weaning or completion of drug regimens at home.

Pediatric hospitalization contributes substantially to the number and type of medications that patients are instructed to take at home after discharge. Clinical decision support and deprescribing programs, which have been shown to have benefits in adult patient populations,^40,41  should be considered to reduce discretionary prescriptions at discharge, particularly among infants, postsurgical patients, and patients who were in the PICU. Opioid prescribing at postsurgical discharge is a high priority. Further research is required to test strategies to reduce overuse and polypharmacy at discharge and determine if such efforts reduce the risks of drug–drug interactions and adverse drug events, improve adherence to essential medications, and lower overall expenditures on drugs in the outpatient setting.

We thank Mark Rullo (Hospital for Sick Children) for compiling the study data set.