In-Hospital Prescription Checking System for Hospitalized Patients with Decreased Glomerular Filtration Rate

Key Points

• We introduced a simple in-hospital prescription checking system to alert for the presence of renally excreted drugs and to support dosage settings.

• The in-hospital prescription checking system reduced the dosage error rate of renally excreted drugs in hospitalized patients.

Visual Abstract

Abstract

Background

Clinical decision support systems (CDSS) are reported to be useful in preventing dosage errors in renally excreted drugs by alerting hospital pharmacists to inadequate dosages for hospitalized patients with decreased GFR. However, it is unclear whether CDSS can reduce dosage errors in renally excreted drugs in hospitalized patients. To prevent dosage errors in renally excreted drugs, we introduced a prescription checking system (PCS) for in-hospital prescriptions. This retrospective study aimed to evaluate whether a prescription audit by hospital pharmacists using the PCS reduced the rate of dosage errors in renally excreted drugs.

Methods

The target drugs were allopurinol, cibenzoline, famotidine, and pilsicainide. Interrupted time series analysis was used to evaluate trends in the 4-weekly dosage error rates over 52 weeks before PCS implementation and 52 weeks after PCS implementation.

Results

Before and after PCS implementation, 474 and 331 prescriptions containing one of the targeted drugs, respectively, were generated. The estimated baseline level of the 4-weekly dosage error rates was 34%. The trend before the PCS implementation was stable with no observable trend. The estimated level change from the last point in the pre-PCS implementation to the first point in the PCS implementation was −20% (P<0.001). There was no change in the trend after PCS implementation.

Conclusions

We demonstrated that a prescription audit by hospital pharmacists using the PCS reduced the rate of dosage errors in the target renally excreted drugs in hospitalized patients. Although further studies are needed to confirm whether our results can be generalized to other health facilities, our findings highlight the need for a PCS to prevent the overdose of renally excreted drugs.

Introduction

CKD has recently been recognized as a public health problem (1,2). CKD is associated with an increased mortality rate and an increased risk of death and cardiovascular events (3). The estimated overall prevalence of CKD in adults (aged ≥20 years) continues to increase in the older population in Japan (4). Patients with CKD often show pharmacokinetic changes (5), and patients with decreased GFR have a greater risk of plasma concentration above the therapeutic range, leading to adverse drug events (ADEs) (6). Corsonello et al. (7) reported that decreased GFR was substantially more prevalent in patients with ADEs from renally excreted drugs. Therefore, adjustment of drug dosages of renally excreted drugs according to kidney function is necessary for patients with decreased GFR (8). The reported overdose rate for renally excreted drugs in hospitalized patients with decreased GFR is 28%–74% (9–11). Thus, the prevention of overdose from renally excreted drugs in hospitalized patients with decreased GFR is an important issue.

Clinical decision support systems (CDSS) are useful in preventing dosage errors in renally excreted drugs by alerting hospital pharmacists to exceeded or contraindicated dosages for hospitalized patients with decreased GFR (12–14). However, previous reports have not evaluated whether CDSS can reduce dosage errors. Using dosage errors as the primary end point, Bhardwaja et al. (15) showed that a prescription audit using CDSS in outpatients at community pharmacies could reduce dosage errors in renally excreted drugs. However, it is unclear whether a prescription audit by hospital pharmacists using CDSS for hospitalized patients would show a similar reduction in dosage errors.

To prevent dosage errors in renally excreted drugs in hospitalized patients, we introduced the following in-hospital prescription checking system (PCS): (1) the label “renal” was added in front of the name of renally excreted drugs on the prescription, (2) the level of the patient’s estimated kidney function was added to the prescription, and (3) a check sheet for dosages according to kidney function was used.

In this study, we retrospectively evaluated whether a prescription audit by hospital pharmacists using the PCS reduced the rate of dosage errors in renally excreted drugs in hospitalized patients.

Materials and Methods

Study Design

We retrospectively analyzed the medical records of hospitalized patients in the Izumi Regional Medical Center. We included patients admitted to the wards of gastrointestinal surgery, neurosurgery, orthopedic surgery, gastrointestinal medicine, ophthalmology, and urology. We conducted a fact-finding survey to identify the renally excreted drugs that tended to be overdosed. Prescriptions containing the renally excreted drugs included in the drug dosing guidelines of the Japanese Society of Nephrology and Pharmacotherapy (16) were examined. The target drugs were determined on the basis of the findings of the survey. A dosage error was defined as a prescription dispensing one of the target drugs at dosages that were contraindicated or too high according to the patient’s kidney function. The following data were collected: age, sex, height, body weight, serum creatinine (SCr) levels, and the prescription contents. In the fact-finding survey, kidney function was estimated using the most recent SCr, age, and body weight data at the time of the prescription audit.

For kidney function estimates, estimated creatinine clearance (eCCr) was obtained using Equation 1 (17):

$$\mathrm{eCCr}\left(\mathrm{ml}/\min \right)=(\text{14}0-\mathrm{age}\left(\mathrm{years}\right))\times \mathrm{body}\mathrm{weight}\left(\mathrm{kg}\right)/(\text{72}\times \mathrm{SCr}\left(\mathrm{mg}/\mathrm{dl}\right))\left(\times 0.85\mathrm{if}\mathrm{female}\right)$$ (1)

The eCCr was automatically calculated using the most recent SCr, age, and body weight data from the in-hospital prescription (Figure 1A).

Figure 1.

The in-hospital prescription checking system. (A) Typical example of an in-hospital prescription containing a renally excreted drug, allopurinol. The label “renal” was added in front of the name of the renally excreted drug and the value of estimated creatinine clearance (eCCr; in milliliters per minute) was added to the prescription. (B) The check sheet of dosages according to kidney function for the target renally excreted drug. The value of eCCr was 27.78 ml/min, and the dosage of allopurinol needed to be reduced from 300 to 50 mg/d. ID, identifier; Rx, prescription; SCr, serum creatinine; Sig, signetur.

The target drug dosages appropriate for different levels of kidney function were determined using a report on allopurinol dosages (18) and the Japanese Society of Nephrology and Pharmacotherapy drug dosing guidelines (16). Dosages of the target drugs according to kidney function are shown in Figure 1B.

This study was approved by the ethics committee of Izumi Regional Medical Center (20170629-1).

Pharmacist Intervention

A workflow diagram of the physician’s order entry, the pharmacist’s prescription audit using the PCS, and the pharmacist’s dispensation is shown in Figure 2. When physicians prescribed the drugs using the computerized physician order entry system, the prescription drug information was sent to the VP-Win total prescription analysis system (TOSHO Inc., Tokyo, Japan). Pharmacists printed the in-hospital prescriptions using the VP-Win system (Figure 1A). If the label “renal” had been added in front of the name of the renally excreted drug on the in-hospital prescription, pharmacists also printed the check sheet of dosages according to kidney function using the VP-Win total prescription analysis system (Figure 1B). Pharmacists checked the dosages of renally excreted drugs using the PCS at the time of dispensation. If the dosage of the target drugs was appropriate, pharmacists dispensed the drugs. If the dosage of the target drugs was inappropriate, pharmacists asked the prescriber about the prescription contents before dispensing the drugs.

Figure 2.

A workflow diagram of the physician’s order entry, the pharmacist’s prescription audit using the prescription checking system (PCS), and the pharmacist’s dispensation. The PCS automatic/electronic processes are enclosed in red boxes.

Interrupted Time Series Analysis with Segmented Linear Regression

A quasi-experimental interrupted time series (ITS) analysis with segmented linear regression was used to evaluate the effect of the PCS (19). ITS analysis has been successfully used to evaluate the effect of various quality improvement programs (20–22). Parameters included trend before PCS implementation (β₁, baseline trend), level change from the last point in the pre-PCS implementation to the first point in the PCS implementation (β₂), and trend change after the PCS implementation (β₃). The regression model was Y_t=β₀+β₁T+β₂D_A+β₃P_A+e_t, where Y_t indicates the 4-weekly dosage error rates, T is the time from baseline, D_A is a dummy variable for the implementation phase (assumes zero before the PCS implementation and one after the PCS implementation), P_A is the period from the PCS implementation, and the error term e_t at time t represents the random variability not explained by the model. The PCS was introduced on August 4, 2016. Therefore, two time segments in the ITS analysis with segmented regression were defined as 52 weeks before the PCS implementation (from August 6, 2015 to August 3, 2016) and 52 weeks after the PCS implementation (from August 4, 2016 to August 2, 2017). The autocorrelation of the residuals in the full segmented regression model was examined using the Durbin–Watson test (23). The periodicity in the 4-weekly dosage error rates of the target drugs was examined using the autocorrelation function (24,25).

Statistical Analysis

Continuous variables are expressed as mean±SD or median (range). The normality of the data was assessed using the Shapiro–Wilk test. Univariate analyses to compare two groups were performed using the Welch t test, Mann–Whitney U test, or Fisher exact test. Significance was set at P<0.05 for all analyses. The statistical analyses were performed using Excel 2010 (Microsoft Corp., Redmond, WA) with the add-in software Ekuseru-Toukei 2012 (Social Survey Research Information Co., Ltd., Tokyo, Japan) and the free software R (version 4.0.4) (26).

Results

Results of the Fact-Finding Survey of Prescriptions Containing Renally Excreted Drugs

Table 1 shows the results of the fact-finding survey of prescriptions containing renally excreted drugs. The number of dosage errors was highest for allopurinol, followed by amantadine and famotidine. Of the drugs with dose errors in three or more cases, the dosage error rates were highest for cibenzoline, pilsicainide, and disopyramide.

Table 1.

Results of the fact-finding survey of prescriptions containing renally excreted drugs from March 2015 to February 2016

Drugs	Prescriptions of Adequate Dose, n	Prescriptions of Dosage Errors, n	Rate of Dosage Errors, %
Allopurinol	151	102	40
Amantadine	81	75	48
Famotidine	262	46	15
Sitagliptin	235	45	16
Metformin	93	37	28
Disopyramide	5	33	87
Risedronate sodium	69	22	24
Levofloxacin	304	17	5
Cibenzoline	11	15	58
Rivaroxaban	214	12	5
Alogliptin	37	9	20
Pilsicainide	8	8	50
Methotrexate	73	8	10
Bezafibrate	51	6	11
Pramipexole	22	5	19
Alogliptin-pioglitazone combination	45	5	10
Entecavir	6	2	25
Tenofovir disoproxil	6	2	25
Edoxaban	85	2	2
Adefovir dipivoxil	0	1	100
Lamivudine	0	1	100
Cimetidine	52	1	2
Pregabalin	468	1	0.2

Patient Characteristics Before and After Implementation of the PCS

Patient characteristics are shown in Table 2. The median ages before and after PCS implementation were 81 and 79 years, respectively. The median SCr levels before and after PCS implementation were 0.86 and 0.86 mg/dl, respectively. The median eCCr (ml/min) before and after PCS implementation were 50.2 and 51.6 ml/min, respectively. There was no significant difference in patient characteristics before and after PCS implementation.

Table 2.

Characteristics of patients before and after implementation of the prescription checking system

Characteristics	System Implementation of the PCS		P Value
	Before	After
Sex, n male/female	84/49	78/52	0.61
Age (yr), median (range)	81 (38–101)	79 (34–102)	0.54
Body weight (kg), mean±SD	56.2 ± 13.0	56.1 ± 13.0	0.93
Serum creatinine (mg/dl), median (range)	0.86 (0.31–8.68)	0.86 (0.24–4.74)	0.14
eCCr (ml/min), median (range)	50.2 (5.2–167.2)	51.6 (5.3–216.9)	0.17
eGFR (ml/min per 1.73 m2), median (range)	56.2 (5.2–141.0)	60.0 (7.1–233.7)	0.14
CKD stage, n (%)
Stage G1	11 (8)	10 (8)
Stage G2	50 (38)	55 (42)
Stage G3a	33 (25)	41 (32)
Stage G3b	20 (15)	17 (13)
Stage G4	15 (11)	5 (4)
Stage G5	4 (3)	2 (2)
Dialysis, n (%)	2 (2)	0 (0)	0.50

PCS, prescription checking system; eCCr, estimated creatinine clearance.

Trend in the 4-Weekly Dosage Error Rates Before and After Implementation of the PCS

Figure 3 shows the trend in the 4-weekly dosage error rates before and after PCS implementation. Table 3 summarizes the parameter estimates from the segmented regression model. The r value for the regression model was 0.90, and the adjusted r² value was 0.79. The estimated baseline level of the 4-weekly dosage error rates (β₀) was 34%. The trend before PCS implementation (β₁) was stable with no observable trend. The estimated level change from the last point in the pre-PCS implementation to the first point in the PCS implementation (β₂) was −20% (P<0.001). There was no trend change after PCS implementation (β₃).

Figure 3.

Change in the 4-weekly rates of dosage errors of the target drugs before implementation of the PCS (52 weeks) and after implementation of the PCS (52 weeks).

Table 3.

Coefficients (95% CIs) and P values from the full segmented linear regression model predicting 4-weekly dosage error rates of target drugs

Full Segmented Regression Model	Coefficient (95% Confidence Interval)	P Value
Intercept, β0a	0.34 (0.26 to 0.41)	<0.001
Trend before the system implementation of the PCS, β1b	−0.01 (−0.02 to 0.00004)	0.05
Level change from last point in the presystem implementation of the PCS to the first point in the system implementation of the PCS, β2c	−0.20 (−0.30 to −0.09)	0.0007
Trend change after the system implementation of the PCS, β3d	0.01 (−0.002 to 0.03)	0.08

PCS, prescription checking system.

^aβ₀ estimates the baseline level of the dosage error rates of the target drugs, at time zero.

^bβ₁ estimates the trend in the 4-weekly dosage error rates of the target drugs before PCS implementation (i.e., the baseline trend).

^cβ₂ estimates the level change in the 4-weekly dosage error rates of the target drugs immediately after PCS implementation.

^dβ₃ estimates the trend change in the 4-weekly dosage error rates of the target drugs after PCS implementation.

The Durbin–Watson statistic for the full segmented regression model predicting the 4-weekly dosage error rates of the target drugs was 1.80 (P=0.12), indicating no autocorrelation. The analysis showed no periodicity in the 4-weekly dosage error rates of the target drugs using the autocorrelation function.

Comparison of the Dosage Error Rate Before and After PCS Implementation

Table 4 shows the comparison of the dosage error rate before and after PCS implementation. The overall dosage error rate of the four target drugs was significantly lower after PCS implementation than before implementation (3% versus 26%, respectively; P<0.001). In particular, the dosage error rate for allopurinol was substantially lower after PCS implementation than before (3% versus 29%, respectively; P<0.001).

Table 4.

Comparison of the dosage error rate of the target drugs before and after implementation of the prescription checking system

Drugs	System Implementation of the PCS				P Valuea
	Before		After
	Prescriptions, n	Prescriptions Containing Dosage Errors, n (%)	Prescriptions, n	Prescriptions Containing Dosage Errors, n (%)
Allopurinol	208	61 (29)	105	3 (3)	<0.001
Cibenzoline	34	18 (53)	14	4 (29)	0.20
Famotidine	207	33 (16)	204	2 (1)	<0.001
Pilsicainide	25	13 (52)	8	2 (25)	0.24
Overall	474	125 (26)	331	11 (3)	<0.001

PCS, prescription checking system.

^aFisher exact test.

Discussion

In this study, we demonstrated that a prescription audit by hospital pharmacists using the PCS reduced the dosage error rate of the target renally excreted drugs in hospitalized patients.

In the fact-finding survey of prescriptions containing renally excreted drugs, the number of dosage errors was highest for allopurinol, followed by amantadine and famotidine (Table 1). In addition, the dosage error rate was highest for antiarrhythmic drugs, such as cibenzoline, pilsicainide, and disopyramide (Table 1). These drugs have frequently been associated with overdose-related ADEs in patients with CKD in Japan (27). These findings suggest that dosage errors frequently occur with high-risk drugs that tend to induce ADEs when overdosed in patients with decreased GFR. Therefore, we recommend the use of interventions to improve this situation. On the basis of these findings, the target drugs selected to test the PCS were allopurinol, cibenzoline, famotidine, and pilsicainide.

Older hospitalized patients frequently have impaired kidney function, despite normal SCr levels, and are exposed to an increased risk of ADEs from renally excreted drugs (7). Therefore, considerable care is needed when administering renally excreted drugs to older hospitalized patients, and adjustment of drug dosages according to kidney function is required. In this study, the median age both before and after PCS implementation was approximately 80 years (Table 2). The median eCCr both before and after PCS implementation was approximately 50 ml/min (Table 2). These results suggest that this study identified a high-risk population for ADEs from renally excreted drugs.

In this study, the ITS analysis showed that the estimated level change from the last point in the pre-PCS implementation to the first point in the PCS implementation was −20% (P<0.001), and the changes in the trend per 4 weeks before and after PCS implementation were −1% and 0.3%, respectively (Figure 3, Table 3). These results suggest the level change was substantially greater than the changes in the trend per 4 weeks before and after PCS implementation. Therefore, we believe the PCS implementation was strongly associated with a reduction in dosage errors. Furthermore, the dosage error rate after PCS implementation declined for all target drugs compared with before PCS implementation (Table 4). These results suggest prescription audit by hospital pharmacists using the PCS can reduce dosage errors in the target drugs.

Our PCS had several advantages over other CDSS. First, the concept underlying our PCS was very simple. The aim was to make pharmacists aware that renally excreted drugs had been prescribed by adding a label to the prescription, and to print out information to facilitate dose determination for patients who were renally impaired. Implementation of this simple concept may be possible regardless of which systems are used by individual hospitals. Second, in our present PCS workflow, pharmacists were always involved in the dosage decisions for the target drugs on the basis of patients’ kidney function. Many previous studies (15,28–30) reported that hospital pharmacists can contribute to dosage setting and reduction in the incidence of ADEs in patients with renal impairment. Therefore, this collaboration between prescribers and pharmacists may have increased the number of correct dosage decisions. Indeed, our findings seem to indicate that our PCS reduced dosage errors compared with other CDSS (31–33) that warn prescribers at the time of prescribing (Table 4). However, because pharmacists are involved in the final dosage decision, they must have a certain level of clinical knowledge about pharmacotherapy for renal impairment. Therefore, pharmacists may need to be well educated about renal impairment before introduction of this PCS system.

This study had some limitations. First, we examined changes in the prescription trend before and after PCS implementation. The number of prescriptions containing allopurinol, cibenzoline, and pilsicainide reduced after the PCS implementation compared with before implementation (Table 4). Therefore, we are unable to exclude the possibility that the prescription trend changes affected the reduction in dosage errors. However, the dosage error rate for famotidine significantly declined after PCS implementation although there was no change in prescription trend (Table 4). In addition, the level change from the last point in the pre-PCS implementation to the first point in the PCS implementation was substantially greater than the changes in the trend per 4 weeks before and after PCS implementation (Table 3). Therefore, we believe that the PCS implementation contributed to the reduction in dosage errors of the target drugs. Second, we were unable to investigate whether the PCS significantly reduced the incidence of ADEs, because the ADE frequency was low in this study. One patient experienced hypoglycemia due to cibenzoline overdose before PCS implementation, but no patients experienced ADEs after PCS implementation. Therefore, we believe that prescription audit by hospital pharmacists using the PCS reduced the dosage error rate of renally excreted drugs, which helped to prevent ADEs after PCS implementation. Third, we cannot rule out the possibility of misidentification of AKI as CKD in the one-point assessment of kidney function in the prescription audit using the PCS. To resolve this issue, we believe it is necessary to devise a way to obtain multiple assessments of kidney function. Indeed, there was one patient with suspected AKI in this study. For this patient, the SCr was elevated before and during the prescription audit, further elevated after the prescription audit, and then returned to near baseline. The pharmacist followed these changes carefully, so there were no problems associated with renally excreted drugs for this patient. The safe use of renally excreted drugs in patients with AKI may require continuous follow-up by medical professionals, with or without PCS. Fourth, there is limited evidence regarding the dosage of renally excreted drugs in AKI, and SCr levels are not at a steady state during AKI; hence, SCr change lags behind both kidney injury and kidney recovery (34–36). Therefore, it is difficult to determine the dosage of renally excreted drugs in the presence of AKI, even with the PCS. Fifth, this study was performed at a single institution, limiting the generalizability of our results. Although further studies are needed to confirm whether our findings can be generalized to other institutions, we believe that PCS implementation in many other institutions could facilitate the appropriate use of renally excreted drugs.

In conclusion, these findings demonstrated that prescription audit by hospital pharmacists using the PCS reduced the dosage error rate of the target renally excreted drugs in hospitalized patients. Although further studies are needed to confirm whether our results can be generalized to other health facilities, our findings highlight the need for PCS implementation to prevent the overdose of renally excreted drugs.

Disclosures

T. Irie reports having patents or royalties with the Safety Medical System Laboratory Corporation. Y. Ishitsuka reports having patents or royalties with the Safety Medical System Laboratory Corporation. Y. Kondo reports receiving research funding from AYUMI Pharmaceutical Corporation and Safety Medical System Laboratory Corporation; having patents or royalties with the Safety Medical System Laboratory Corporation; and serving in an advisory or leadership role for Safety Medical System Laboratory Corporation (unpaid). All remaining authors have nothing to disclose.

Funding

None.

Author Contributions

T. Irie, K. Ishida, Y. Ishitsuka, Y. Iwashita, Y. Kondo, S. Nakao, and A. Sonoda were responsible for methodology; T. Irie, Y. Ishitsuka, and Y. Kondo were responsible for funding acquisition and provided supervision; T. Irie. Y. Ishitsuka¸ Y. Kondo, and A. Sonoda reviewed and edited the manuscript, and were responsible for project administration and visualization; Y. Ishitsuka, Y. Kondo, and A. Sonoda were responsible for investigation; Y. Kondo and A. Sonoda wrote the original draft and were responsible for data curation and formal analysis; A. Sonoda was responsible for resources; and all authors conceptualized the study.

Data Sharing Statement

The datasets analyzed in this study are available from the corresponding author on reasonable request.