Abstract
Background:
Adverse drug events (ADEs) result in excess hospitalizations. Thorough admission medication histories (AMHs) may prevent ADEs; however, the resources required oftentimes outweigh what is available in large hospital settings. Previous risk prediction models embedded into the Electronic Medical Record (EMR) have been used at hospitals to aid in targeting delivery of scarce resources.
Objective:
To determine if an AMH scoring tool used to allocate resources can decrease 30-day hospital readmissions.
Design, Setting, and Participants:
Propensity-matched cohort study, Medicine/Surgery patients in large academic safety-net hospital.
Intervention or Exposure:
Pharmacy-conducted AMHs identified by risk model versus standard of care AMH.
Main Outcomes and Measures:
A total of 30-day hospital readmissions and inpatient ADE prevention.
Results:
The model screened 87 240 hospitalizations between June 2017 and June 2019 and 4027 patients per group were included. There were significantly less 30 day readmissions among high-risk identified patients that received a pharmacy-conducted AMH compared to controls (11% vs 15%; P = 0.004) and no significant difference in readmission rates for low-risk patients. While there was significantly higher documentation of major ADE prevention in the pharmacy-led AMH group versus control (1656 vs 12; P < 0.001), there was no difference in electronically-detected inpatient ADEs between groups.
Conclusions:
A risk tool embedded into the EMR can be used to identify patients whom pharmacy teams can easily target for AMHs. This study showed significant reductions in readmissions for patients identified as high-risk. However, the same benefit in readmissions was not seen in those identified at low-risk, which supports allocating resources to those that will benefit the most.
Keywords: adverse drug reactions, clinical pharmacy services, medication therapy management, medication safety, monitoring drug therapy, technicians
Purpose
To determine if an AMH scoring tool used to allocate pharmacy resources can decrease 30-day hospital readmissions and inpatient adverse drug events.
Background
The results from a survey conducted by the Boston University found 80% of Americans responded to taking at least 1 medication within the last week and one-third responded to taking more than 5 within the last week. 1 An adverse drug event (ADE) is defined as any harm or injury that results from medication use. 2 Each year in the United States, ADEs account for more than 1 million visits to hospital emergency departments (ED) resulting in over $3.5 billion being spent on excess medical costs.1,3 It is estimated that 27% of ED visits for adverse drug events resulted in hospitalizations. The most commonly implicated drug classes are anticoagulants/antiplatelets, antibiotics, antidiabetic medicines, opioid analgesics. 4
Identified determinants of preventable ADEs include poor health literacy, polypharmacy, age, use of high-risk medications, and medical and/or psychiatric comorbidities. 5 Polypharmacy is of particular concern among elderly patients who may see multiple specialists for various comorbidities. 6 Borhanjoo et al found that the number of patient comorbidities was a significant predictor of 30-day readmissions and that each added comorbidity significantly increased 30-day readmission rates by 26%. 7 Patients are especially vulnerable to medication discrepancies and medication errors during transitions of care. 8 Medication discrepancies have been reported to account for over half of medication errors therefore, medication reconciliation has been identified as a major intervention to target. 7 Previous studies that include a multi-pronged approach involving pharmacists in transitions of care, including medication reconciliation, have shown a reduction in ADEs; however, they have shown mixed results in hospital readmissions.7,9-12
Previous readmission risk prediction models embedded into the electronic medical record (EMR) have been used at hospitals to aid in targeting delivery of scarce resources and timeliness of interventions.13,14 Given this information, identifying high risk populations and specific elements of medication reconciliation services may aid in allocating resources while simultaneously lowering readmissions rates and ADEs.
In efforts to improve timely and accurate medication reconciliations, a pharmacy admission medication history team was formed at Parkland Health and Hospital System. The team consists of two pharmacy technicians and one pharmacist. The pharmacy team has the capacity to complete medication histories on 6% to 8% of admissions, proper identification of patients at risk for adverse drug events allows for better allocation of resources. Therefore, a risk scoring tool named PARADE (Patients at risk for Adverse Drug Events), was incorporated to proactively identify the top 10% of patients at high risk for ADEs that would benefit from timely admission medication histories by a pharmacy team. The risk scoring tool consists of a weighted score for high risk medications, comorbid disease states and social factors. Risk factors and their weights are integrated into the EMR, allowing for dynamic scoring and ranking of all patients.
The purpose of this research is to determine if the implementation of an admission medication history (AMH) scoring tool for resource allocation at a large academic safety-net institution can be used to identify patients at high risk for adverse drug events and lead to decreased 30-day hospital readmissions.
Methods
Study Design
The study was a single center, propensity-matched cohort study conducted at a large academic safety-net hospital.
Population
Medicine and Surgery patients 18 years of age and older that were admitted to Parkland Hospital between June 2017 and June 2019 were identified for inclusion.
Workflow
Patients admitted through the emergency department were identified for pharmacy-led admission medication histories using the risk tool; however, provider activated consults were also included. The risk factor synopsis is available in the EMR patient lists and can be used by the pharmacy team for detailed information on score make up (Figure 1). Prior to implementation of the risk tool, it took 10 minutes per patient to screen for risk level and intervention.
Figure 1.
Patient lists in electronic medical record used to identify which patients to prioritize for pharmacy-led admission medication histories.
A team of 2 pharmacy technicians and 1 pharmacist complete medication histories and clinical interventions on patients who were either consulted to them by a provider or were chosen by risk scoring. The service is offered between the hours of 8:00am and 6:30pm. Once a patient is identified, the pharmacy technician gathers the medication history by interviewing the patient and/or caregiver, reviews medication lists or bottles, contacts nursing homes or retail/specialty/mail order pharmacies, accesses state-controlled substance databases and health information exchange portals available in the EMR. The prior-to-admission (PTA) medication list is updated in the EMR and discrepancies between how the patient is supposed to take versus how they are taking the medication are documented. Translation services are used if English is not the primary language for the patient and/or caregiver. All medications, including OTC and non-U.S. available products are included in the final medication list and a note is written in the chart. In addition to gathering the medication list, the following are documented: allergies, side effects related to non-adherence, and refills needed. After the pharmacy technician completes the admission medication history both the treatment team and the clinical pharmacist are notified.
A clinical pharmacist reviews the medication list to identify medication related problems and discrepancies between home medications versus those started in the hospital. After the review is complete, the pharmacist writes a consult note in the EMR and notifies the provider with recommendations. Some additional services provided include: disposal of old or expired medications, providing copay assistance cards, scheduling of pharmacist post discharge follow-up visits, and providing medication adherence tools such as pill boxes and syringe magnifiers.
Pharmacy team interventions were captured using the intervention documentation tool in the EMR to indicate if a major or minor ADE was prevented based on pharmacist judgement.
Model Development
The PARADE model was developed in a retrospective, cohort study of 55 000 adult hospitalizations. This real-time algorithm screens every newly admitted adult patient in the hospital to identify those at highest risk for a potential preventable ADE and can benefit from pharmacy intervention. It generates a real-time actionable worklist within the EMR for pharmacy team.
It conceptually captures medications, disease complexity, prior healthcare utilization, demographics, and social determinants of health. While several machine learning models were trained and tested, multivariate logistic regression was chosen for having a better model performance and clinical relevance.
The PARADE model’s multifaceted risk factors and their weights are integrated into Epic’s Patient Acuity Scoring, allowing user-friendly display of the risk scores in green-yellow-red color palette. The colors red (96th-100th percentile), yellow (91th-95th percentile) and green (10th-90th percentile) are associated with very high, high and low risk patients for ADE respectively. Contributing risk factors are mined through chart and made available to end user as a snapshot adjacent to the score/color in the same view. Seamless EMR integration enables dynamically ranked worklists and chart synopsis for timely pharmacy team interventions.
Score Components
After an extensive literature driven research; inputs from subject matter experts and end users like clinical pharmacists; consideration of patient safety reports/manuals; mining of the EMR to shortlist multiple medications which potentially could result in ADEs were performed. Given all these inputs, feature selection via machine learning algorithms to identify the significant features for the model was used to finalize the list of medications that had shown better performance for the model and also which had a clear clinical relevance.
The model incorporated the medications insulin, phenytoin, opioids, clozapine, metolazone, anticoagulants, and antihyperglycemics. Each medication category was mapped to an objective measure for a potential or actual ADE (Table 1). Disease states incorporated include: diabetes mellitus, chronic lung disease, heart failure, renal failure, liver disease, and substance abuse. Other criteria the model included were age, number of hospitalizations and ED visits, payer status, non-English speaking, and from where the patient was being admitted (e.g., Skilled Nursing Facility). One logistic regression model predicted the presence of a consult and provided weights to the selected binary covariates. The same covariates were used in a second logistic regression model to predict the ADEs (potential and occurred). The weights provided by consult and ADE models were reconciled with clinical input by stakeholders to push performance closer to best practice. On the training dataset, the consult model has c-statistic 0.68. When applied directly to predict any ADE, it has a c-statistic of 0.64. The logistic regression model is able to provide fine-tuned weights to the riskiness of individual predictors. For example, while previously thought to be equivalent, prior Heart Failure and Liver disease were stronger predictors of ADE then prior Renal/Lung disease, and Diabetes was the least predictive. Odds ratios from ADE model correlated well with the odds ratios from the consult model which validates that providers initiate consults in high risk patients. The slight discrepancies were adjusted with clinical input and the final weights were implemented. Patients with scores in the top 10% based on historical data are identified as high risk.
Table 1.
List of Potential or Actual Adverse Drug Event Mapping from High Risk Medication to Objective EMR Data.
| Drug category | Adverse drug outcome (actual or potenial) | Results/lab a | ICD10 a |
|---|---|---|---|
| Anticoagulants | Bleeding | X | |
| Warfarin | Supra/subtherapeutic levels | X | |
| Phenytoin | Supra/subtherapeutic levels | X | |
| Phenytoin | Seizure | X | |
| Metolazone | Hypokalemia | X | |
| Opioids | Respiratory depression | X | |
| Clozapine | Neutropenia | X | |
| Methadone | QTc prolongation | X | |
| Antihyperglycemics | Hypoglycemia | X |
Different time frames were given based on t1/2 of the drug.
Validation of PARADE
The PARADE prediction model was trained and tested based on a 50:50 data split of Parkland retrospective patient encounters of the 2016 calendar year. The model was validated using the test set during the development phase and the model output was further evaluated for clinical relevance. During the post development phase, the model performance was further re-validated using prospective clinical encounters for a 90-day pilot run. The results were compared to the previous model development performance to ensure similarity. Additionally, a thorough clinical validation through chart reviews were performed to understand and interpret the model output as compared to a clinical judgement on the patient’s risk for an ADE. After successful validation of pilot run, the PARADE model was implemented in production environment for end user availability. Furthermore, post deployment periodic evaluation was conducted to ensure optimal model performance and clinical significance. The current manuscript offers a 2-year model performance evaluation of clinical outcomes.
Outcomes Analysis
The primary outcome was the rate of 30-day all-cause readmission between those patients with and without a pharmacy-team admission medication history completion by risk score. The secondary outcomes included the number of pharmacist documented ADE prevention interventions and electronically captured potential or actual ADEs. Primary and secondary outcomes were compared using the chi-square test and an alpha of 0.05 to indicate statistical significance.
Baseline characteristics were compared using standardized differences as explained by Austin. 15 One of the commonly used methods to handle biases and confounding factors in observational studies or randomized clinical trials has been propensity score matching where probability of a subject being in treatment or control group is determined.16,17 Propensity Score matching was also used to achieve a balance of covariates in treatment and control groups.
Logistic regression was trained based on the baseline covariates as independent variables. Backward stepwise elimination was used to eliminate insignificant features. The final model had 14 of the baseline covariates as the variables that were significant for determining the propensity score (excluding race, ethnicity, primary diagnosis, length of stay). Propensity scores were determined using this model where propensity score is a probability of a patient getting treated or intervened upon. After this step, one-to-one matching for each patient in the treatment group with a patient of similar propensity score within the neighborhood of 0.1 times the standard deviation of propensity score from the control group was done (0.1 was chosen to ensure that there were matches for each patient in our case). In case of multiple matches based on a propensity score, randomized selection from the multiple matches was used to select the patient into control group. The differences in readmission rates and adverse drug events were compared statistically using chi-square test.
Results
After the model was incorporated into the EMR, there were 87 240 patient encounters from Parkland Hospital between June 2017 and June 2019 that were identified for inclusion. Baseline characteristics for treated and untreated patients in the original unmatched sample by hospital encounter are shown in Table 2. Overall, 4027 patients received a pharmacy team AMH while the remaining 83 213 patients did not receive a consultation due to feasibility and limitation of resources. Significant differences were found in most of the demographic, disease state, medication, and utilization variables prior to propensity-score matching.
Table 2.
Comparison of Baseline Characteristics Between Treated and Untreated Subjects in the Original Unmatched Sample by Hospital Encounter.
| Pharmacy team AMH N = 4027 | Control N = 83 213 | P -value | Std Diff | |
|---|---|---|---|---|
| Mean age (SD) | 56.5 (15.1) | 50.34 (16.0) | <.001† | 0.395 |
| Female (%) | 1713 (42.5%) | 39225 (47.1%) | <.001 ‡ | 0.094 |
| Co morbidities (%) | ||||
| Diabetes mellitus | 2110 (52.1%) | 37468 (45.0%) | <.001 ‡ | 0.143 |
| Chronic lung disease | 264 (6.5%) | 9279 (11.2%) | <.001 ‡ | 0.194 |
| Renal failure | 2094 (51.7%) | 33450 (40.2%) | <.001 ‡ | 0.265 |
| Heart failure | 1621 (40.1%) | 20276 (24.4%) | <.001 ‡ | 0.341 |
| Liver disease | 710 (17.6%) | 17358 (20.9%) | <.001 ‡ | 0.115 |
| Avg # of ED/UC Visits prior 12 mo | 1.9 (5.9) | 2.1 (6.9) | .04 † | −0.031 |
| Avg # of hospitalizations prior 12 mo | 1.3 (2.3) | 1.2 (2.4) | .01 † | 0.043 |
| Avg # of PTA meds | 5.5 (3.5) | 4.97 (3.4) | <.001 † | 0.154 |
| High risk/ high alert medications* on PTA | 328 (8.1%) | 3899 (4.6%) | <.001 ‡ | 0.14 |
| Length of stay (in days) | 6.7 (9.7) | 5.1 (7.8) | <.001 † | 0.182 |
| Model risk score cat | <.001 ‡ | 0.647 | ||
| High risk (top 10%) | 1389 (34.4%) | 7286 (8.8%) | ||
| Low risk | 2638 (65.5%) | 72655 (90.8%) | ||
| Mean model score | 1.53 (0.8) | 1.13 (0.5) | <.001 † | 0.599 |
| Primary payor | <.001 ‡ | 0.436 | ||
| Medicare | 1475 (36.6%) | 19433 (24.4%) | 0.397 | |
| Medicaid | 963 (23.9%) | 15336 (19.2%) | 0.009 | |
| Charity | 865 (21.5%) | 22971 (28.8%) | 0.168 | |
| Self-pay | 466 (11.6%) | 15371 (19.3%) | 0.213 | |
| Others | 258 (6.4%) | 6636 (8.3%) | 0.081 | |
| Jail as payor | 358 (8.8%) | 654 (0.8%) | <.001 ‡ | 0.147 |
| Race (%) | <.001 ‡ | 0.295 | ||
| Asian | 70 (1.7%) | 1813 (2.2%) | ||
| African American | 1889 (46.8%) | 27189 (32.6%) | ||
| White | 2019 (50.1%) | 52863 (63.5%) | ||
| Other | 49 (1.2%) | 1348 (1.5%) | ||
| Ethnicity (%) | <.001 ‡ | 0.412 | ||
| Hispanic | 1031 (25.6%) | 37311 (44.8%) | ||
| Non-Hispanic | 2965 (73.6%) | 42265 (54.4%) | ||
| Unknown | 31 (0.8%) | 633 (0.8%) | ||
| Discharge Dx | ||||
| Circulatory (%) | 966 (23.9%) | 14885 (18.6%) | <.001 ‡ | 0.131 |
| Cerebrovascular | 30 | 1414 | ||
| Heart Failure | 395 | 3257 | ||
| Inflammatory | 5 | 179 | ||
| Digestive (%) | 435 (10.8%) | 11696 (14.6%) | <.001 ‡ | 0.115 |
| Liver | 69 | 863 | ||
| Alcohol-related | 94 | 2186 | ||
| Endo/metabolic (%) | 421 (10.5%) | 5996 (7.5%) | .006 ‡ | 0.103 |
| Diabetes | 246 | 3573 | ||
| Genitourinary (%) | 176 (4.3%) | 3646 (4.5%) | .59 ‡ | 0.009 |
| Renal | 164 | 2848 | ||
| Infectious diseases (%) | 625 (15.5%) | 13135 (16.4%) | .13 ‡ | 0.024 |
| Pulmonary (%) | 267 (6.6%) | 4478 (5.6%) | .006 ‡ | 0.043 |
| Injury/ingestions (%) | 308 (7.6%) | 6753 (8.5%) | .08 ‡ | 0.029 |
| Neoplasms (%) | 61 (1.5%) | 4149 (5.2%) | <.001 ‡ | 0.205 |
| Musculoskeletal/connective tissue (%) | 83 (2.1%) | 1975 (2.5%) | .11 ‡ | 0.027 |
| Nervous system/sense organs (%) | 163 (4.0%) | 3259 (4.1%) | .96 ‡ | 0.001 |
| Other (%) | 522 (13%) | 9969 (12.5%) | .37 ‡ | 0.0147 |
ED/UC = Emergency Department/Urgent Care; PTA = prior to admission.
t-test.
Chi-squared test.
Antihyperglycemic agents, anticoagulant agents, opioids, clozapine, metolazone, and phenytoin.
After propensity-score matching, the number of patients in both the new control group and treatment group was 4027 each. Table 3 displays the baseline characteristics for the treated and untreated patients in the propensity-score matched sample by hospital encounter. The standardized differences are presented to check the balance of the matched cohorts in order to compare the prevalence and means of the baseline covariates. Table 3 shows a reduction in the differences of baseline characteristics between groups in categories such as age, comorbidities, when compared to the differences of baseline characteristics before matching in Table 2.
Table 3.
Comparison of Baseline Characteristics Between Treated and Untreated Subjects in the Propensity-Score Matched Sample by Hospital Encounter.
| Pharm team AMH N = 4027 | Control N = 4027 | Std Diff | |
|---|---|---|---|
| Mean age (SD) | 56.5 (15.1) | 55.3 (15.1) | 0.08 |
| Female (%) | 1713 (42.5%) | 1731 (43.0%) | 0.009 |
| Co morbidities | |||
| Diabetes mellitus | 2110 (52.1%) | 2113 (52.4%) | 0.001 |
| Chronic lung disease | 264 (6.5%) | 278 (6.9%) | 0.014 |
| Renal failure | 2094 (51.7%) | 2096 (52.0%) | 0.001 |
| Heart failure | 1621 (40.1%) | 1669 (41.4%) | 0.024 |
| Liver disease | 710 (17.6%) | 754 (18.7%) | 0.028 |
| Avg # of ED/UC visits prior 12 mon | 1.9 (5.9) | 2.1 (6.6) | 0.029 |
| Avg # of hospitalizations prior 12 mon | 1.3 (2.3) | 1.4 (2.4) | 0.069 |
| Avg # of PTA meds | 5.5 (3.5) | 5.6 (3.6) | 0.028 |
| High risk/high alert medications* on PTA | 328 (8.1%) | 353 (8.7%) | 0.069 |
| Length of stay (in days) | 6.7 (9.7) | 5.4 (8.5) | 0.179 |
| Model risk score cat | 0.012 | ||
| High risk (Top 10%) | 1389 (34.4%) | 1366 (33.9%) | |
| Low | 2638 (65.5%) | 2661 (66.1%) | |
| Mean model score | 1.53 (0.8) | 1.50 (0.8) | |
| Primary payor | |||
| Medicare | 1475 (36.6%) | 958 (23.8%) | 0.248 |
| Medicaid | 963 (23.9%) | 1017 (25.3%) | 0.003 |
| Charity | 865 (21.5%) | 1274 (31.6 %) | 0.232 |
| Self-Pay | 466 (11.6%) | 485 (12.0%) | 0.015 |
| Others | 258 (6.4%) | 293 (7.3%) | 0.034 |
| Inmate as the payor | 358 (8.8%) | 292 (7.3%) | 0.144 |
| Race (%) | 0.217 | ||
| Asian | 70 (1.7%) | 75 (1.9%) | |
| African American | 1889 (46.8%) | 1472 (36.6%) | |
| White | 2019 (50.1%) | 2436 (60.2%) | |
| Other | 49 (1.2%) | 44 (1.1%) | |
| Ethnicity (%) | 0.318 | ||
| Hispanic | 1031 (25.6%) | 1625 (40.3%) | |
| Non-hispanic | 2965 (73.6%) | 2374 (58.9%) | |
| Unknown | 31 (0.8%) | 28 (0.6%) | |
| Discharge Dx | |||
| Circulatory (%) | 966 (23.9%) | 1085 (26.9%) | −0.068 |
| Cerebrovascular | 30 | 94 | |
| Heart failure | 395 | 332 | |
| Inflammatory | 5 | 14 | |
| Digestive (%) | 435 (10.8%) | 494 (12.3%) | −0.046 |
| Liver | 69 | 39 | |
| Alcohol-related | 94 | 71 | |
| Endo/metabolic (%) | 421 (10.5%) | 317 (7.9%) | 0.089 |
| Diabetes | 246 | 192 | |
| Genitourinary (%) | 176 (4.3%) | 181 (4.5%) | −0.006 |
| Renal | 164 | 155 | |
| Infectious diseases (%) | 625 (15.5%) | 626 (15.5%) | −0.001 |
| Pulmonary (%) | 267 (6.6%) | 209 (5.2%) | 0.061 |
| Injury/ingestions/medications (%) | 308 (7.6%) | 330 (8.2%) | −0.020 |
| Neoplasms (%) | 61 (1.5%) | 147 (3.7%) | −0.134 |
| Musculoskeletal/connective tissue (%) | 83 (2.1%) | 80 (2%) | 0.005 |
| Nervous system/sense organs (%) | 163 (4.0%) | 126 (3.1%) | 0.049 |
| Other (%) | 522 (13%) | 432 (10.7%) | 0.069 |
Antihyperglycemic agents, anticoagulant agents, opioids, clozapine, metolazone, and phenytoin.
In Figure 2, no significant differences in readmission rates were observed in the propensity-score matched treatment or control groups when the sample included all patients (composite high-risk and low-risk patients) or low-risk patients alone. When the high-risk groups alone were compared, there was a significant reduction in 30-day readmission rates compared to the control group (11% vs 15% respectively; P = 0.004).
Figure 2.
Primary outcome of 30-day all cause readmission in patients with pharmacy team admission medication history versus standard of care.
The pharmacy team documented significantly more minor and major ADE preventions in the pharmacy-led AMH group. However, there was no statistically significant difference in selected electronically captured major ADEs between the matched pharmacy-led AMH group compared to the control group (Table 4).
Table 4.
Adverse Drug Events and Prevention Documentation by Pharmacy Team.
| Medications—ADEs | Treatment |
Control |
p -value | ||||
|---|---|---|---|---|---|---|---|
| # | N* | % | # | N* | % | ||
| Pharmacist documented ADE prevention | |||||||
| Minor ADE prevented | 1656 | 4027 | 41.1 | 12 | 4027 | 0.2 | <.001 |
| Major ADE prevented | 88 | 4027 | 2.2 | 3 | 4027 | 0.001 | <.001 |
| Electronically detected ADE | |||||||
| Bleeding—Anticoagulant agent | 11 | 306 | 3.6 | 9 | 283 | 3.1 | .96 |
| INR out of range—Warfarin | 62 | 170 | 36.4 | 62 | 170 | 36.4 | 1 |
| Respiratory depression—Opioid agent | 7 | 184 | 3.8 | 9 | 174 | 5.1 | .73 |
| Hypoglycemia—Anti-hyperglycemic Agent | 30 | 180 | 16.6 | 16 | 180 | 8.8 | .07 |
| Hypokalemia—Metolazone | 0 | 18 | 0 | 2 | 33 | 6 | .78 |
| Drug level out of range—Phenytoin | 9 | 30 | 30 | 7 | 32 | 21.9 | .78 |
| Seizure—Phenytoin | 1 | 30 | 3.3 | 3 | 32 | 9.4 | .69 |
Note. *Eligible patients qualifying for analysis.
The scoring system allowed for faster identification of high risk patients by front-line care teams in their daily workflows and resulted in high risk patients being 5 times more likely to receive a pharmacy consult. It is estimated that 0.16 FTE/year were saved as a result of real-time identification of patients when compared to manual identification.
Discussion
The MARQUIS study, which has been the largest medication reconciliation improvement study to date highlights the challenges of implementing real world interventions such as allocation of resources and workflow modifications. 18 Previous studies have hypothesized that it may be more resource effective to use risk prediction models to decrease readmissions.13,19,20 We found this to be true in our study as well, as the same benefit was not seen for patients categorized as low risk for ADEs. This study demonstrates that the use of a validated risk tool for identifying patients at highest risk for ADEs allows us to allocate scarce pharmacy resources and decrease 30-day readmissions. The scoring system allows for faster identification of high risk patients by front-line care teams in their daily workflows which reduces time spent identifying these patients by 0.16 FTE/year.
Overall, significantly more adverse drug event preventions were documented in patients receiving a pharmacy-team intervention. Unintended medication discrepancies occur 38 times more often when there is no clinical pharmacist intervention and the likelihood for experiencing discrepancies at admission increases by over 47% for each additional medication listed in a medication history.21,22 This reaffirms the importance of having pharmacist involvement in the medication reconciliation process, especially for complicated medication regimens where their expertise is warranted.
Despite significantly more documentation of ADEs prevented by the pharmacy team, our study did not show a reduction in select inpatient ADEs using objective lab and ICD10 data. This finding coupled with the positive impact on 30-day readmissions could mean that the interventions pharmacists made were more impactful on the accuracy and changes to the home medication list, including refill recommendations, rather than clinical interventions that would change ADEs occurring during the initial hospitalization.
Another potential contributing factor may be low to moderate acceptance rates of clinical recommendations provided by AMH pharmacists, which was not collected as part of this study. 23 Anderegg et al found that only about 50% of inpatient pharmacist recommendations were accepted in their study. Physicians were more likely to accept recommendations for a record update (P < 0.001) and less likely to accept recommendations regarding drug indication (P < 0.001) and efficacy (P = 0.041). Physicians accepted less than 50% of recommendations for potential ADEs and ADRs, untreated or undertreated conditions, monitoring for efficacy and just over half of the recommendations for inappropriate or suboptimal doses. Possible reasons that may have contributed to low acceptance rates include that pharmacists provided recommendations on medical conditions outside of the scope of the primary reason for hospital admission and that pharmacists were not as well known to physicians since they did not directly round with teams.
A major strength of this study is that the PARADE risk tool was developed and trained in a retrospective study of 55 000 adult hospitalizations and was validated on 3 months of prospective encounters prior to its use in this study. In addition, propensity score matching was used to account for differences in diabetes mellitus, renal failure, heart failure, high risk PTA medications, length of stay, Medicaid patients in pre-matched group that occurred due to natural workflow favoring higher risk patients. Propensity score methods have been well described in the literature as a tool to reduce bias in observational studies. 15
A major limitation to our study is that the model has only been validated at one medical institution. Weights for risk of individual variables will change between institutions and should be considered. We would further hope to get to better performance when the model is retrained to have additional variables from literature and further clinical input for its deployment, if any. The model is also subjective to several biases. First, only a select list of ADEs are examined. Second, the severity of each ADE is not equivalent. Third, not all inpatient ADEs are related to or impacted by home medication list and they are not all equally preventable by timely pharmacy team involvement. Nevertheless, this is a quantitative approach that significantly improves the current process and brings it closer to reflecting the clinical need.
Conclusion
A validated risk tool embedded into the EMR can be used to identify patients that pharmacy teams can easily target for AMHs. This study showed significant reductions in readmissions for patients identified as high-risk. However, the same benefit in readmissions was not seen in those identified as low risk, which supports allocating resources to those that will benefit the most. Furthermore, the PARADE risk tool has the potential to be modified for use at other institutions after accounting for differences in patient populations and input variables.
Acknowledgments
The authors would like to thank Lanora Gray, CPhT, Parkland Health and Hospital System and Genae Sam, CPhT, Parkland Health and Hospital System for their guidance and support.
Footnotes
Author Contributions: Hanh Nguyen: Writing—Original draft preparation, Investigation, Writing—Review & Editing. Kristin Alvarez: Conceptualization, Methodology, Validation, Investigation, Resources, Writing—Review & Editing, Visualization, Supervision, Project administration. Boryana Manz: Software, Methodology, Validation, Investigation, Data Curation, Writing—Review & Editing. Arun Nethi: Methodology, Validation, Formal analysis, Investigation, Data Curation, Writing—Review & Editing. Varun Sharma: Investigation, Resources, Validation. Venkatraghavan Sundaram: Writing—Review & Editing. Manjula Julka: Conceptualization, Methodology, Validation, Investigation, Resources, Writing—Review & Editing, Visualization, Supervision, Project administration.
Declaration of Conflicting Interests: The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding: The author(s) received no financial support for the research, authorship, and/or publication of this article.
ORCID iD: Hanh L. Nguyen 
https://orcid.org/0000-0003-4154-9561
References
- 1. Slone Epidemiology Center at Boston University. Patterns of medication use in the United States. 2006. Updated 2006. Accessed February 7, 2018. http://www.bu.edu/slone/files/2012/11/SloneSurveyReport2006.pdf
- 2. Bates DW, Cullen DJ, Laird N, et al. Incidence of adverse drug events and potential adverse drug events. Implications for prevention. ADE Prevention Study Group. JAMA. 1995;274(1): 29-34. [PubMed] [Google Scholar]
- 3. Aspden P, Wolcott J, Bootman JL, Cronenwett LR. Committee on Identifying and Preventing Medication Errors: Preventing Medication Errors. Institute of Medicine National Academies Press; 2006. Accessed February 1, 2018. https://www.nap.edu/initiative/committee-on-identifying-and-preventing-medication-errors. [Google Scholar]
- 4. Shehab N, Lovegrove MC, Geller AI, Rose KO, Weidle NJ, Budnitz DS. US emergency department visits for outpatient adverse drug events, 2013-2014. JAMA. 2016;316(20):2115-2125. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. U.S. Department of Health and Human Services, Office of Disease Prevention and Health Promotion. National Action Plan for Adverse Drug Event Prevention. Washington, DC; 2014. [Google Scholar]
- 6. Shrestha S, Shrestha S, Khanal S. Polypharmacy in elderly cancer patients: Challenges and the way clinical pharmacists can contribute in resource-limited settings. Aging Med (Milton). 2019;2(1):42-49. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Borhanjoo P, Kouamo P, Rahman M, Norton M, Gavini M. Effect of clinical pharmacist encounters in the transitional care clinic on 30-day re-admissions: a retrospective study. AIMS Public Health. 2019;6(3):345-354. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Rozich JD, Howard RJ, Justeson JM, Macken PD, Lindsay ME, Resar RK. Standardization as a mechanism to improve safety in health care. Jt Comm J Qual Saf. 2004;30(1):5-14. [DOI] [PubMed] [Google Scholar]
- 9. Baker M, Bell CM, Xiong W, et al. Do combined pharmacist and prescriber efforts on medication reconciliation reduce postdischarge patient emergency department visits and hospital readmissions? J Hosp Med. 2018;13(3):152-157. [DOI] [PubMed] [Google Scholar]
- 10. Phatak A, Prusi R, Ward B, et al. Impact of pharmacist involvement in the transitional care of high-risk patients through medication reconciliation, medication education, and postdischarge call-backs (IPITCH Study). J Hosp Med. 2016;11(1):39-44. [DOI] [PubMed] [Google Scholar]
- 11. Shull MT, Braitman LE, Stites SD, Deluca A, Hauser D. Effects of a pharmacist-driven intervention program on hospital readmissions. Am J Health Syst Pharm. 2018;75(9):e221-e230. [DOI] [PubMed] [Google Scholar]
- 12. Gardella JE, Cardwell TB, Nnadi M. Improving medication safety with accurate preadmission medication lists and postdischarge education. Jt Comm J Qual Patient Saf. 2012;38(10):452-458. [DOI] [PubMed] [Google Scholar]
- 13. Kansagara D, Englander H, Salanitro A, et al. Risk prediction models for hospital readmission: a systematic review. JAMA. 2011;306(15):1688-1698. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Amarasingham R, Patel PC, Toto K, et al. Allocating scarce resources in real-time to reduce heart failure readmissions: a prospective, controlled study. BMJ Qual Saf. 2013;22(12): 998-1005. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Austin PC. Balance diagnostics for comparing the distribution of baseline covariates between treatment groups in propensity-score matched samples. Stat Med. 2009;28(25): 3083-3107. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Xu Z, Kalbfleisch JD. Propensity score matching in randomized clinical trials. Biometrics. 2010;66(3):813-823. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Burden A, Roche N, Miglio C, et al. An evaluation of exact matching and propensity score methods as applied in a comparative effectiveness study of inhaled corticosteroids in asthma. Pragmat Obs Res. 2017;8:15-30. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Schnipper JL, Mixon A, Stein J, et al. Effects of a multifaceted medication reconciliation quality improvement intervention on patient safety: final results of the MARQUIS study. BMJ Qual Saf. 2018;27(12):954-964. [DOI] [PubMed] [Google Scholar]
- 19. D’Agostino R., Jr. Propensity score methods for bias reduction in the comparison of a treatment to a non-randomized control group. Stat Med. 1998;17(19):2265-2281. [DOI] [PubMed] [Google Scholar]
- 20. Burke RE, Coleman EA. Interventions to decrease hospital readmissions: keys for cost-effectiveness. JAMA Intern Med. 2013;173:695-698. [DOI] [PubMed] [Google Scholar]
- 21. Leguelinel-blache G, Arnaud F, Bouvet S, et al. Impact of admission medication reconciliation performed by clinical pharmacists on medication safety. Eur J Intern Med. 2014;25(9):808-814. [DOI] [PubMed] [Google Scholar]
- 22. Cornu P, Steurbaut S, Leysen T, et al. Effect of medication reconciliation at hospital admission on medication discrepancies during hospitalization and at discharge for geriatric patients. Ann Pharmacother. 2012;46(4):484-494. [DOI] [PubMed] [Google Scholar]
- 23. Anderegg SV, Demik DE, Carter BL, et al. Acceptance of recommendations by inpatient pharmacy case managers: unintended consequences of hospitalist and specialist care. Pharmacotherapy. 2013;33(1):11-21. [DOI] [PMC free article] [PubMed] [Google Scholar]