Cost-effectiveness of two technology-assisted manual medication picking systems versus traditional manual picking in a hospital outpatient pharmacy

Abstract

Objective

To evaluate the cost-effectiveness of two technology assisted manual medication picking systems vs traditional manual picking.

Methods

This was a retrospective observational study comparing three outpatient pharmacies of a tertiary referral hospital in Singapore, where a light-emitting diode (LED-guided) manual picking system, an LED-guided manual picking plus lockable drawer (LED-LD) system, and traditional manual picking were implemented, respectively. The primary outcome measure was the incidence of medication near-misses over the observation period. The incremental cost-effectiveness ratio (ICER) per near-miss avoided was also determined. Data on medications picked and near-misses reported between September 2017 and June 2018 were retrieved from electronic databases. The incidence of medication near-misses from the LED-guided and LED-LD systems, relative to traditional picking, was compared using logistic regression. We compared annual operating costs between manual medication picking systems, and reported ICERs per near-miss avoided, to evaluate the cost-effectiveness of each picking system.

Results

A total of 358 144, 397 343 and 254 162 medications were picked by traditional manual picking, LED-guided and LED-LD systems, respectively. The corresponding near-miss rates were 8.32, 4.08 and 0.69 per 10 000 medications picked, respectively. Medication near-miss rates were significantly lower for the LED-guided (OR 0.49, 95% CI 0.40 to 0.59, p<0.001) and LED-LD systems (OR 0.08, 95% CI 0.05 to 0.13, p<0.001) compared with traditional picking. The annual operating costs of traditional picking, LED-guided and LED-LD systems were S$60 912, S$129 832 and S$152 894, respectively. The LED-guided and LED-LD systems yielded ICERs of S$189 and S$140 per near-miss avoided, respectively, compared with traditional manual picking.

Conclusion

The LED-LD system is more cost-effective than both the LED-guided and manual medication picking systems in reducing medication picking near-misses.

Introduction

Medication safety is an important issue for healthcare institutions. It is imperative to be cognisant of potential consequences of a medication error, which can range from the need for additional monitoring to requiring medical or surgical intervention, or even death.¹ The Institute of Medicine estimated that medication errors cause 1 out of every 131 outpatient deaths.² Medication errors have been associated with increased morbidity and mortality, contributing significantly to healthcare costs, while compromising patients’ confidence in the healthcare system.³ In the US, costs associated with medication errors have been estimated to range between US$2.8 million to almost US$15 million per year at academic medical centres, amounting to extrapolated nationwide costs of over US$2 billion annually.⁴

Nonetheless, medication errors only represent a small percentage of reported error events. A prospective study of 35 community pharmacies in the United Kingdom reported that dispensing medication errors accounted for only 15.2% of reported errors.⁵ The remaining 84.8% of reported errors were considered “near-misses”, defined by the Institute for Safe Medication Practices as errors that have the potential to cause harm but that do not reach the patient (due to chance, or because they are intercepted).⁶ Near-misses and medication errors often share the same root causes.⁷ Therefore, interventions aimed at reducing near-misses are crucial as they are also likely to prevent future medication errors. The implementation of automated systems such as medication picking carousels has been identified as a means to reduce medication errors and near-misses in pharmacies.^8–10 This includes a reduction in dispensing errors, which account for 8%–17.8% of all medication errors.^11–13 Fitzpatrick et al reported a 45% reduction in medication picking errors, an increase in prescription processing efficiency, and a reduction in dispensing time by 19% following the implementation of an automated dispensing system to replace traditional picking stations.¹⁴

However, a concurrent manual picking component running parallel to the pharmacy automation system is often necessary. Bulky, odd shaped or low-usage items often cannot be handled by automated systems, and backup systems are often required during automation downtime and maintenance.¹⁵ These concurrent manual workflows pose a potential source of medication errors. Cousein et al reported that 18.3% of medications still needed to be manually picked and loaded with automated pharmacy machineries, which contributed to the 4% error rate following implementation of the Pillpick and Pyxis systems.¹⁶ In addition, the implementation of pharmacy automation solutions also adds to the cost of running the pharmacy, incuding infrastructure, land and additional staff training costs.¹⁷

In an effort to minimise medication errors associated with traditional manual picking, the outpatient pharmacies of Singapore General Hospital (SGH) implemented two different technologies to facilitate manual picking of medications – the light-emitting diode (LED-guided) picking system and the LED-guided picking plus lockable drawer (LED-LD) system. We sought to assess whether these systems are effective in reducing medication near-misses, and evaluated the cost-effectiveness and affordability of both systems compared with traditional manual picking – which might be useful in aiding pharmacy budgetary decisions.

Methods

Study design and setting

We conducted a retrospective observational study using electronic data from three outpatient pharmacies in SGH (an 1800-bed tertiary referral hospital) between September 2017 and June 2018. These pharmacies have a combined inventory size of 7500 medications and dispense a total of 1500 prescriptions daily. The LED-guided and LED-LD systems were implemented to facilitate manual medication picking at two of the three pharmacies in 2012 and 2016, respectively. The third pharmacy uses traditional manual picking stations and is the comparator in our study. Medications stored in refrigerators and controlled drug cabinets were excluded since medication picking systems were not implemented in these areas.

Description of manual medication picking systems

Traditional system

Traditional manual picking involves the use of open access medication bins, labelled with location tags and located on the shelving units of the medication carousels. Medications are picked by pharmacy staff based on the location code indicated on the medication label, which corresponds to the medication bin location.

LED-guided and LED-guided plus lockable drawer (LED-LD) systems

The LED-guided and LED-LD systems are linked to our medicine management information system, allowing extraction of data on the location of the medication. Both systems have adopted LED technology to aid in locating the correct medication bin. On scanning the Quick Response (QR) code of a medication label, the LED corresponding to the medication location lights up, directing the pharmacy staff to the correct medication bin. (figure 1)

Figure 1.

Schematic of the ‘LED-guided manual picking with lockable drawer’ system, the ‘LED-guided manual picking system’ and the traditional manual picking system.

The LED-guided system utilises open access medication bins lined with LED strips. In contrast, the LED-LD system utilises a remote lockable drawer system, integrated with LED technology: on scanning the QR code, only the drawer containing the corresponding medication is unlocked, enabling its retrieval; the drawer locks automatically when it is shut.

Near-miss reporting and data collection

As the interventions are specifically aimed at reducing errors arising from the medication picking stage of prescription processing, our study focused solely on medication picking-related near-misses. We defined “near-miss” as any picking error that has the potential to cause harm, but that is detected before being supplied to the patient during dispensing. In our institution, medication near-misses are reported on a voluntary basis using a standardised self-reporting database. Standardised briefings were provided for all staff to encourage accurate and standardised near-miss reporting.

Medication errors only encompass a small percentage of incidents that could potentially lead to adverse events, leading to a small database that might not be adequate for data analysis. Our study focused on near-misses because a larger database is available for such potential errors. This allowed us to perform comparisons that would not have been possible with the adverse event reporting system.

All medications picked during the study period were reviewed. Anonymized data on medication near-misses (owing to the wrong medications being picked) were retrieved electronically. We included three categories of near-misses that can be overcome with the picking systems: wrong drug, wrong strength, and wrong dosage form. The definitions for each category are detailed in table 1.

Table 1.

Definitions of medication picking-related near-misses

Type of near miss	Definition
Wrong drug	The identity of the medication picked deviated from the medication label.
Wrong strength	The identity of the medication picked matched that of the label, but the strength of the medication deviated from the medication label.
Wrong dosage form	The identity of the medication picked matched that of the label, but the dosage form of the medication deviated from the medication label.

Statistical analysis

Sample size was estimated based on the primary outcome measure, which is the incidence of medication near-misses over a 10-month period. Based on historical data from traditional manual picking, we anticipated a baseline near-miss rate of 8 per 10 000 medications picked. In order to detect a 50% reduction in medication near misses after implementing the medication picking systems with 80% statistical power, 59 000 picking jobs in each group were required.

Descriptive statistics were used to summarise medication near-miss rates in each group. Near-miss rate was calculated as the total number of near-misses divided by the total number of medications picked. We compared the incidence of medication near misses in LED-guided and LED-LD systems with that for traditional manual picking, using binary logistic regression. In addition, we stratified the analyses by the types of medication near-miss, namely those involving picking of the wrong drug, wrong strength or wrong dosage form. All analyses were conducted using Stata 13 (College Station, TX: StataCorp LP) at the p<0.05 significance level.

Economic evaluation

Cost-effectiveness analysis was performed from the institution perspective, using the odds ratios (OR) derived from the logistic regression models as the effectiveness parameter. We applied the OR to the near-miss rate of traditional manual picking to derive the number of near-misses in the LED-guided and LED-LD systems within the 10-month study period, an approach similarly implemented by Risør et al.¹⁸ This was done instead of using the number of near-misses reported in each system so that costs and outcomes could be compared under similar workloads. A near-miss reporting rate of 50% was assumed to account for potential under-reporting of near-misses.

We calculated technological infrastructure and associated software costs using a standard annuitization method to attain the estimated annual cost (EAC) of each picking system, with an annual discount rate of 5%.¹⁹ To account for differences in the volume of medications processed between pharmacies, cost calculations were based on a standardised load across the three pharmacies, adjusted to the pharmacy with the highest load. The lifespan of the technologies was set at 15 years to account for extended lifespan from maintenance. The average annual maintenance cost was added to EAC to account for the total annual cost of the LED-guided and LED-LD systems. We assumed no technological infrastructure cost was incurred in traditional manual picking.

We estimated the annuitized cost of hardware to support the LED-guided and LED-LD systems. We included the costs of desktop computers, barcode scanners linked to each computer for scanning of QR codes, and label printers for generating medication labels at each picking station. The total lease price over the effective lifespan was used as the total cost for resources on lease, and cost price was used for resources purchased. No hardware cost was incurred in traditional manual picking.

The annuitized cost of facilities included construction and land lease costs. Floor area required by the medication shelves, including additional space required for housing the technological hardware, was measured. We multiplied the total facility land area by 1.4 to account for movement space. The required floor area was multiplied by cost per unit area, as stated in our institution’s asset list. Lease cost per annum taken up by facilities was added to the total facilities cost. We obtained annuitized facility costs by sharing total facility cost over an annuity of 15 years.

A time-motion study was performed to account for possible differences in medication picking times between each picking mode. The handling time for each system was multiplied by the hourly wage of pharmacy staff to ascertain the manpower costs incurred for each medication picking system. The annual manpower cost incurred for medication picking was estimated from the total number of medications picked in 1 year.

Finally, we compared the total annual cost and the total near-misses in each picking system, and reported the incremental cost-effectiveness ratio (ICER) per near-miss avoided. We calculated the ICERs using the lower and upper limits of the 95% CI of the ORs obtained from the binary logistic regression model. Deterministic sensitivity analysis was performed to assess uncertainty in the ORs used to estimate the near-miss rates. In addition, we performed subgroup analyses by estimating the ICER for each type of near-miss.

Results

Between September 2017 and June 2018, a total of 1 009 649 medications were picked at the three pharmacies, of which 358 144, 397 343 and 254 162 items were contributed by traditional manual picking, LED-guided and LED-LD systems, respectively. A total of 298, 162 and 17 near misses were reported in the traditional manual picking, LED-guided and LED-LD systems, respectively. This corresponded to a near-miss incidence of 8.32, 4.08 and 0.69 per 10 000 medications picked for the traditional manual picking, LED-guided and LED-LD systems, respectively. As compared with traditional manual picking, the incidence of medication near-miss was significantly lower in the LED-guided manual picking system (OR 0.49, 95% CI 0.40 to 0.59, p<0.001) and LED-LD system (OR 0.08, 95% CI 0.05 to 0.13, p<0.001) (table 2). The results were consistent across all three categories of near-miss, with ORs ranging from 0.20 to 0.62 and 0.07 to 0.11 for the LED-guided and LED-LD systems, respectively, compared with traditional manual picking.

Table 2.

Medication near-misses reported in the LED-guided manual picking and LED-guided manual picking with lockable drawer (LED-LD) systems, compared with traditional manual picking

Type of near-miss	Manual picking system	Near-miss, n	Near-miss rate, no. per 10 000 items	OR	95%	P value
All	Traditional	298	8.32	Reference
	LED-guided	162	4.08	0.49	0.40 to 0.59	<0.001
	LED-LD	17	0.69	0.08	0.05 to 0.13	<0.001
Wrong drug	Traditional	218	6.09	Reference
	LED-guided	149	3.75	0.62	0.50 to 0.76	<0.001
	LED-LD	11	0.43	0.07	0.04 to 0.13	<0.001
Wrong strength	Traditional	54	1.51	Reference
	LED-guided	12	0.30	0.20	0.11 to 0.37	<0.001
	LED-LD	4	0.16	0.10	0.04 to 0.29	<0.001
Wrong dosage form	Traditional	26	0.73	Reference
	LED-guided	1	0.03	0.35	0.004 to 0.26	0.001
	LED-LD	2	0.08	0.11	0.03 to 0.46	0.002

Cost-Effectiveness analysis

The hardware and manpower costs for each picking system are summarised in table 3. The annuitized costs for the traditional manual picking, LED-guided and LED-LD systems amounted to S$60 912, S$129 832 and S$152 894, respectively, with an assumed annuitized volume of 476 812 items. Hardware setup, maintenance and manpower costs accounted for the largest proportion of annuitized cost. As the assumed annuitized volume only takes up 38% of the full capacity of the pharmacies, a further increase in volume will likely increase manpower cost, while facilities and hardware costs remain unchanged.

Table 3.

Annual cost of medication picking systems (assumed annuitized volume of 476 812 items)

Cost component	Cost, S$
	Traditional	LED-guided	LED-LD
Facilities
Rental	16 722	19 753	18 186
Construction	7739	9142	8417
Hardware
Setup and maintenance	0	49 263	65 504
Computers	0	905	905
QR code scanners	0	1134	1133
Printers	0	4073	4073
Manpower
Medication picking	36 451	45 564	54 677
Total annual cost	60 912	129 832	152 894

Compared with traditional manual picking, the LED-guided manual picking system resulted in the avoidance of 365 medication near-misses, corresponding to an ICER of S$189 per near-miss avoided. The LED-LD system resulted in the avoidance of 658 medication near misses, corresponding to an ICER of S$140 per near miss avoided (table 4). However, the LED-guided manual picking system was dominated by the LED-LD system by extended dominance when the comparison was made based on the next best non-dominated system. The ICERs pertaining to avoidance of wrong drug, wrong strength and wrong dosage form near-misses were S$346, S$663 and S$1723 respectively for the LED-guided system, and S$189, S$786 and S$1672 respectively for the LED-LD system, compared with traditional manual picking. Sensitivity analyses showed that ICER did not fluctuate substantially from the base case ICER (table 4). Subgroup analyses of the ICER by type of near-miss yielded similar findings (online supplementary appendix I).

Table 4.

Cost-effectiveness analysis of medication picking systems (assumed annuitized volume of 476 812 items)

Medication picking system	Cost, S$	Incremental cost, S$	Near-misses, n	Incremental reduction in near-misses, n	Incremental cost-effectiveness ratio (range of ICER*), S$/ near-miss avoided
					Compared with traditional	Compared with next best non-dominated system
Traditional	60 912	–	715	–	–	–
LED-guided	129 832	68 920	350	365	189 (160 – 235)	Dominated
LED-LD	152 894	91 982	57	293	140 (136 – 148)	140

*Estimated using the upper and lower limits of the 95% CI of the ORs for medication near-miss.

Discussion

We found that the use of technology for aiding the manual medication picking process was effective in reducing medication near-misses in a hospital outpatient pharmacy. Our study is the first of its kind to assess the cost-effectiveness of using technology to improve manual medication picking processes in an outpatient pharmacy. The use of a common study period mitigated potential confounders such as time-dependent differences in pharmacy workload due to outbreaks or hectic periods. The large number of prescriptions filled at our outpatient pharmacies provided sufficient statistical power, despite the low incidence of medication near-misses reported.

The implementation of technology-assisted medication picking systems resulted in a reduction of more than 50% in the number of medication near-misses. This was consistent with studies from James et al and Fanning et al, which found 56% and 64.7% reductions in near-miss rates after implementation of automated medication dispensing systems, respectively.^{20 21} These results indicate that technology-assisted medication picking systems are effective in improving medication safety.

Our study reported medication near-miss rates of 8.32, 4.08 and 0.69 per 10 000 medications for the traditional manual picking, LED-guided and LED-LD systems, respectively, which was lower than the estimated 165 to 210 per 10 000 medications reported by the European Medicines Agency and by other studies.^{20 22} This could have been attributed to the broader definition of near-miss used by other studies, which encompassed other aspects such as incorrect labelling, assembly errors and administration errors. Our study only included errors that technology-assisted medication picking systems are anticipated to have a greater impact on.²³

Given the increased costs that accompany the implementation of new technologies, cost-effectiveness determination is crucial for decision-making. Our pharmacy incurred an additional S$189 (US$139) and S$140 (US$103) for every near-miss avoided using the LED-guided and LED-LD systems, respectively, when compared with traditional manual picking. These ICERs are higher than the ICER of US$22.17 per clinical error avoided that Risør et al reported for an electronic pharmacy ordering system coupled with an automated medication picking system.¹⁸ Thise disparity might be due to differences in practice settings and in the computation of costs. While some parallels may be drawn between studies conducted in the inpatient and outpatient settings, these studies may not be directly comparable due to differences in technology and study design. Furthermore, technologies employed in the inpatient setting (such as automated medication cabinets) have more automated elements and hence a lower propensity for human error, resulting in larger effect sizes and lower costs per medication error avoided. Consequently, we were unable to establish meaningful comparisons of ICER thresholds, and recognise this as a limitation of our study.

Our findings suggest that the LED-LD system is more cost-effective compared with both the LED-guided and manual medication picking systems. The presence of the self-locking mechanism in the LED-LD system prevented staff from overriding procedures and retrieving medications without following the LED indicator. This could have contributed to the 92% reduction in the odds of a medication near-miss occurring in the LED-LD system compared with traditional manual picking, driving the ICER in favour of the LED-LD system. Moreover, the annual cost of the LED-LD systems account for a small proportion of our institution’s annual medication budget (S$10 million in 2016), making it an affordable investment from the institutional perspective. A lower near-miss rate could also result in other forms of cost savings, such as the manpower cost incurred to rectify the near-miss. Lastly, as the LED-LD system is highly scalable, our findings may be applicable to other outpatient pharmacies looking to adopt similar technologies to reduce potential medication errors.

Our study has limitations. First, differences in pharmacy practices and voluntary reporting of errors could have confounded the findings. Nonetheless, we anticipate this to be minimal as the same error reporting platform was used throughout the three pharmacies, and all staff received standardised training in its use. In addition, the error reporting platform is embedded in the dispensing interface, minimising variation in platform accessibility. Second, we were unable to rule out effects of time differences, since the LED-guided and LED-LD systems were implemented in 2012 and 2016, respectively. However, potential effects may have been mitigated by standardised staff training to operate and troubleshoot systems, and by on-site technical support. Moreover, most setup issues for the LED-LD system were resolved within 1 year following implementation. Lastly, the type of medication dispensed may vary between pharmacies due to differences in inventory holding, resulting in bias in assessing near-miss rates. However, we anticipate the effect to be small as there was substantial overlap (>50%) in the types of medication available at each pharmacy.

Conclusion

Our findings suggest that technology-assisted medication picking systems are effective in reducing medication near-misses, but at a higher cost. The cost-effectiveness analysis suggests that the LED-LD system is more cost-effective compared with both the LED-guided and manual medication picking systems. It remains an affordable option given the associated low proportion of budgetary consumption for our institution. The findings will be useful to hospital administrators looking to implement similar technology to improve medication safety.

Key messages.

What is already known on this subject

• Pharmacy automation and robotics have an important role in reducing operational inefficiency and medication picking errors.

• Manual picking is often necessary alongside automated medication picking to cater for items that cannot be picked via machines and as a machine-downtime contingency.

• More often than not, medication picking error rate is increased with manual picking, due to human error.

What this study adds

• Technologies that serve to increase medication safety during manual medication picking are available.

• Cost-effectiveness studies of these strategies are lacking, but are crucial for stakeholder decision-making during technology set-up.

• This study introduces the ‘LED-guided with lockable drawer system’ as an alternative strategy to reduce errors during manual medication picking, and to evaluate its cost- effectiveness vs traditional manual picking in a hospital outpatient pharmacy.

• The findings will be useful to hospital administrators looking to implement technology to improve safety in their medication picking processes.