Effect of a Medication Decision Support System on Albumin Prescribing: A Pre-post Study in Intensive Care Units of Imam Reza Hospital, Mashhad, Iran

Abstract

Objective:

Medication Decision Support Systems (MDSS) are increasingly integrated into hospital information systems to reduce prescribing errors and enhance evidence-based clinical decision-making. This study evaluates the effect of MDSS implementation on albumin prescribing in two intensive care units (ICUs) at Imam Reza Hospital, Mashhad, Iran.

Methods:

A quasi-experimental pre-post design was applied. Albumin prescription data were collected over two 3-month phases, before and after MDSS implementation. Total population sampling was used. Statistical analyses included Chi-square and independent-sample t-tests to assess differences in guideline adherence, alert responsiveness, and patient safety. P < 0.05 was considered significant, and 95% confidence intervals were reported where applicable.

Findings:

A total of 311 albumin prescription requests were reviewed. Following MDSS implementation, 60.15% of alerts led to prescription modification. Guideline adherence improved significantly (from 47.64% to 68.26%, P = 0.014), and patient safety rates increased (63.33% to 82.61%, P = 0.009). Alert responsiveness was highest in critical conditions such as acute respiratory distress syndrome and lowest in elective scenarios such as paracentesis.

Conclusion:

MDSS significantly improved guideline adherence and prescribing quality in ICU settings. However, system limitations such as alert fatigue and physician override in specific scenarios remain. Further research is warranted to evaluate MDSS scalability, long-term clinical impact, and application to broader drug categories.

INTRODUCTION

Clinical decision-making in intensive care units (ICUs) involves complex pharmacologic interventions and high-risk medications. Among these, albumin is frequently prescribed for indications such as shock, liver cirrhosis, hypoalbuminemia, and nephrotic syndrome. However, multiple studies have demonstrated that albumin is often used outside of guideline-based indications, leading to medication errors, patient risk, and increased healthcare costs.[1,2,3]

In Iran, research has shown significant variation in albumin prescribing practices and a high rate of rejections by Drug Utilization Evaluation (DUE) committees due to off-label or undocumented use.[4] These trends underscore the need for structured systems that reinforce guideline adherence and minimize unnecessary prescriptions of high-cost medications.[3,5]

To address these challenges, Medication Decision Support Systems (MDSS) – a subtype of Clinical Decision Support Systems (CDSS) – have emerged as valuable tools that provide evidence-based alerts and recommendations at the point of care.[6,7,8] These systems can integrate clinical guidelines, patient-specific data, and hospital policies to reduce medication-related errors and optimize therapy. For instance, Kaushal et al. and Kuperman et al. reported that computerized physician order entry (CPOE) systems with embedded MDSS modules significantly improved prescribing accuracy and reduced error rates.[8,9]

However, MDSS’s impact is not uniformly positive. Some studies have reported conflicting results, including poor physician compliance with alerts, limited effect on outcomes, and alert fatigue – where clinicians override or ignore system prompts.[10,11] These limitations highlight the importance of evaluating MDSS performance in local, real-world settings to understand contextual factors influencing effectiveness.

In Iran, MDSS research has mainly focused on antimicrobial stewardship programs,[12,13] and relatively few studies have examined its application for nonantibiotic medications such as albumin. There is a lack of evidence regarding clinicians’ real-time behavioral responses to MDSS alerts and the system’s actual impact on improving DUE approvals and patient safety metrics.

This study was conducted in two ICUs at Imam Reza Hospital, Mashhad, Iran, one of the country’s major academic referral centers. Albumin usage in these ICUs was among the highest across hospital wards, making them suitable targets for evaluating MDSS impact. The MDSS was integrated into the hospital information system (HIS), enabling real-time alerts based on institutional guidelines for albumin use.

The specific objective of this study was to assess whether implementing an MDSS could enhance adherence to institutional guidelines, reduce inappropriate albumin use, improve patient safety, and increase physician responsiveness to alerts. A pre- and post-intervention design was employed to compare prescribing behavior and safety outcomes before and after MDSS integration in two ICUs.

The strength of this study lies in its real-world application within a hospital setting, where the MDSS was fully integrated into the HIS and used in routine clinical workflows. The study measured changes in adherence and safety metrics and evaluated physician interaction with system alerts, offering practical insight into clinical behavior and alert acceptance. These features increase the external validity of the findings and support the broader applicability of MDSS tools across other medications and care settings.

METHODS

This study used a quasi-experimental pre-post design to evaluate the effect of a MDSS on albumin prescribing. The research was conducted in two ICUs include Common Intensive Care Unit (ICUC) and Internal Intensive Care Unit (ICUD) of Imam Reza Hospital, a tertiary care academic center affiliated with Mashhad University of Medical Sciences in Mashhad, Iran. These ICUs were selected based on their historically high consumption of albumin. While this focused setting strengthens internal consistency, it may limit the generalizability of the results to other hospital contexts.

The study included all albumin prescription requests submitted during two 3-month periods: Pre-intervention: May 22 to August 22, 2023, and post-intervention: October 23, 2023 to January 20, 2024. Total population sampling was used, meaning all albumin prescriptions in the target ICUs during the study periods were included without sampling. This method eliminated selection bias and ensured complete coverage of prescribing behaviors in both periods. Emergency or undocumented prescriptions (e.g., handwritten without MDSS registration) were excluded from the analysis.

The MDSS module was integrated into the HIS to guide albumin prescribing. It compared each prescription against local and internationally adopted guidelines. When deviations occurred, the system generated real-time alerts prompting prescribers to revise or justify the request. The system automatically logged alert types, physician responses (accepted or overridden), and the final decision.

The MDSS was built using a multilayered model-view-controller architectural design with a centralized drug repository, ensuring that prescribing rules could be updated dynamically. The platform employed a rule-based inference engine, which allowed clinicians or pharmacists to define or modify prescription guidelines without requiring direct programming. To tailor its prescribing recommendations, the system retrieved real-time patient data, including laboratory values, hemodynamic parameters, and concurrent medications. Figure 1 illustrates the principal workflow diagram of albumin prescription and evaluation through the medication decision support system.

Figure 1.

Workflow diagram of Medication Decision Support Systems integration with hospital information systems

Preintervention data were extracted from manual DUE request forms, and postintervention data were automatically captured by the MDSS. Key variables included clinical indication, number of albumin vials prescribed, ICU unit (C or D), presence of MDSS alerts, physician response to alerts, DUE approval or rejection, and patient safety outcomes.

Statistical analysis was performed using IBM SPSS Statistics for Windows, Version 26.0 (IBM Corp., Armonk, NY, USA). Descriptive statistics summarized prescription characteristics. Chi-square tests were used for categorical variables (e.g., adherence rates and alert response), whereas independent sample t-tests compared continuous outcomes (e.g., number of vials). P < 0.05 was considered statistically significant, and 95% confidence intervals (CIs) were reported for primary outcome comparisons. While multivariable modeling was not performed, confounding was partially controlled by analyzing each ICU separately, stratifying results by indication and clinical context, and comparing physician response rates based on alert types.

Due to automated MDSS logging in the postintervention phase, missing data were minimal. Any incomplete records in the preintervention phase (e.g., forms missing indication or dosage) were excluded. No data imputation techniques were applied.

The study was conducted per institutional and national ethical guidelines and was approved by the Ethics Committee of Mashhad University of Medical Sciences (Approval ID: IR.MUMS.REC.1399.990947). All data were anonymized and used solely for research purposes.

RESULTS

A total of 544 patients received albumin prescriptions in the two ICU wards (ICU C and ICU D) during both study phases. Of these, 311 prescription requests were formally submitted to the hospital’s DUE unit for review. The analysis focuses on changes in prescribing patterns, guideline adherence, physician responses to MDSS alerts, and patient safety outcomes before and after MDSS implementation. Detailed data are presented in Tables 1, 2 and Figures 2–4.

Table 1.

Albumin prescription patterns: pre- and post-intervention

Ward name	Registered albumin in HIS	Unregistered albumin in HIS	Submitted requested albumin to DUE	Confirmed requested albumin by DUE	Guideline-base requested albumin	Requested albumin with patient safety	Patient safety rate	Requested vials of submitted requests to DUE	Requested vials of unsubmitted requests to DUE
Preintervention phase
ICUC	113	2	53	39	27	39	73.58	100	116
ICUD	135	14	97	78	43	56	57.73	203	116
Total	248	16	150	117	70	95	63.33	303	232
Average patient safety rate (%)		65.66	Average guideline base rate (%)			47.64	Average confirmed rate (%)		77.00
Postintervention phase
ICUC	115	4	81	66	64	73	90.12	154	105
ICUD	118	3	80	53	46	60	75.00	159	69
Total	273	7	161	119	110	133	82.61	313	174
Average patient safety rate (%)		82.56	Average guideline base rate (%)			68.26	Average confirmed rate (%)		73.87

HIS=Hospital information system, DUE=Drug utilization evaluation, ICUC=Common intensive care unit, ICUD=Internal intensive care unit

Table 2.

Distribution of albumin prescriptions by indication

Before MDSS implementation (pre-intervention)
Indication name	Registered albumin in HIS	Unregistered albumin in HIS	Submitted requested albumin to DUE	Unsubmitted requested albumin to DUE	Confirmed requested albumin by DUE	Unconfirmed requested albumin by DUE	Guideline-base requested albumin	No-guideline-base requested albumin	Requested albumin with patient safety	Requested albumin without patient safety	Patient safety rate	No patient safety rate
ARDS	61	0	21	40	18	3	10	11	13	8	61.90	38.10
Heart failure	17	1	12	6	9	3	6	6	9	3	75.00	25.00
Heart surgery	2	0	2	0	2	0	2	0	2	0	100.0	0.0
Hepatorenal syndrome	66	7	43	30	39	4	19	24	19	24	44.19	55.81
Major surgery	16	2	11	7	4	7	4	7	9	2	81.82	18.18
Nephrotic syndrome	16	0	16	0	15	1	12	4	13	3	81.25	18.75
Organ transplantation	3	0	3	0	2	1	2	1	1	2	33.33	66.67
Cirrhosis of the liver with refractory ascites	22	4	16	10	12	4	6	10	10	6	52.50	37.50
Therapeutic plasmapheresis	2	1	3	0	3	0	3	0	3	0	100.0	0.0
No indication	43	1	23	21	13	10	6	17	16	7	69.57	30.43
Total	248	16	150	114	117	33	70	80	95	55	63.33	36.67
Average patient safety rate (%)			70.96
Average guideline base rate (%)			58.34
Average confirmed rate (%)			77.97
After MDSS implementation (post-intervention)
Indication name	Registered albumin in HIS	Unregistered albumin in HIS	Submitted requested albumin to DUE	Unsubmitted requested albumin to DUE	Confirmed requested albumin by DUE	Unconfirmed requested albumin by DUE	Guideline-base requested albumin	No-guideline-base requested albumin	Requested albumin with patient safety	Requested albumin without patient safety	Patient safety rate	No patient safety rate
ARDS	48	0	33	15	23	10	28	5	28	5	84.85	15.15
Heart failure	35	1	17	19	10	7	14	3	11	6	64.71	35.29
Heart surgery	1	0	1	0	1	0	1	0	1	0	100.0	0.0
Hepatorenal syndrome	6	0	5	1	2	3	2	3	5	0	100.0	0.0
Major surgery	11	0	8	3	6	2	3	5	5	3	62.5	37.5
Nephrotic syndrome	39	3	27	15	20	7	22	5	25	2	92.59	7.41
Cirrhosis of the liver with refractory ascites	33	2	17	18	14	3	11	6	14	3	82.35	17.65
Therapeutic plasmapheresis	1	1	1	1	1	0	1	0	1	0	100.0	0.0
Hemorrhagic shock	3	0	2	1	1	1	0	2	1	1	50.0	50.0
Hypovolemia/sepsis	7	0	5	2	5	0	4	1	4	1	80.0	20.0
Intracranial hemorrhage dialysis treatment in the presence of severe abnormalities of hemostasis	13	0	13	0	12	1	12	1	11	2	84.62	15.38
Paracentesis	14	0	8	6	8	0	8	0	8	0	100.0	0.0
Pleural effusion	14	0	12	2	5	7	8	4	7	5	58.33	41.67
Spontaneous bacterial peritonitis	5	0	4	1	4	0	4	0	4	0	100.0	0.0
No indication	43	0	8	35	7	1	7	1	8	0	100.0	0.0
Total	273	7	161	119	119	42	125	36	133	28	82.61	17.39
Average patient safety rate (%)			84.00
Average guideline base rate (%)			74.49
Average confirmed rate (%)			78.09

HIS=Hospital information system, DUE=Drug utilization evaluation, MDSS=Medication decision support system, ARDS=Acute respiratory distress syndrome

Figure 2.

Guideline adherence rate: pre- and post-intervention

Figure 4.

Patient safety rate: pre- and post-intervention

Figure 3.

Confirmed requests rate: pre- and post-intervention

Before MDSS implementation, a total of 264 albumin prescription requests were generated, with 248 being registered in the hospital’s HIS. Only 150 requests (56.8%) were forwarded to the DUE unit for formal assessment, corresponding to 303 vials of albumin 20%. Notably, 84.66% of DUE-reviewed prescriptions had a documented indication for albumin use.

After MDSS deployment, the total number of albumin requests increased slightly to 280, with 273 (97.5%) formally recorded in the HIS. However, the number of unregistered albumin requests declined from 16 to 7, indicating improved system-based documentation and form completion. Among these, 161 requests (57.5%) were submitted for DUE review, corresponding to 313 albumin vials. More importantly, the completeness of prescribing information improved, with 95% of submitted requests including a documented indication for albumin use.

Guideline adherence improved substantially following MDSS implementation. Before the intervention, only 47.64% of prescriptions reviewed by the DUE unit met established hospital and external guidelines, with ICU C demonstrating a slightly higher adherence rate (50.94%) than ICU D (44.33%). Among the specific indications for albumin therapy, adherence rates varied significantly, with conditions such as heart surgery and plasmapheresis achieving full (100%) compliance. In comparison, major surgery exhibited one of the lowest adherence rates (36.36%).

Following MDSS implementation, guideline adherence increased to 68.26% (P = 0.014, 95% CI: 5.2%–35.9%). ICU C demonstrated a notable improvement, reaching an adherence rate of 79.01%, while ICU D improved to 57.50%. Indications such as paracentesis, bacterial peritonitis, heart surgery, and plasmapheresis reached 100% compliance. These improvements suggest that the MDSS effectively reinforced standardized prescribing behavior, minimizing variability and deviations from best practices.

The impact of MDSS on formal DUE assessments was mixed. Before its deployment, the DUE unit evaluated 150 albumin prescriptions, approving 77% for administration. ICU C prescriptions had a confirmation rate of 73.58%, while ICU D had a higher approval rate of 80.41%. The mean confirmation rate across all indications stood at 77.97%, with heart surgery and plasmapheresis maintaining 100% approval.

Postintervention, 161 albumin prescriptions were reviewed by the DUE unit, with an average confirmation rate of 73.87%. The approval rate for ICU C prescriptions improved to 81.48%, while ICU D’s confirmation rate decreased to 66.25%. This difference may be due to higher physician responsiveness to alerts in ICU C, more override behavior, or reduced documentation in ICU D. Across various indications, the mean confirmation rate remained stable at 78.09%. Interestingly, despite the overall improvements in guideline adherence, specific indications, such as hepatorenal syndrome (HRS), experienced higher rejection rates, with up to 60% of requests denied; this may reflect a mismatch between national guidelines and clinical reality or physician reluctance to delay albumin in high-severity cases, despite system alerts. Further analysis of overrides in HRS suggests clinicians often justified use manually, citing advanced liver failure or diuretic resistance. This highlights the need for more nuanced alert customization for complex cases.

Patient safety, defined as the proportion of prescriptions that adhered to guidelines and received DUE approval, improved significantly following MDSS implementation. Before the intervention, only 63.33% of prescriptions met these conditions. ICU C had a higher patient safety rate (73.58%) than ICU D (57.73%).

Following the implementation of MDSS, overall patient safety increased to 82.61% (P = 0.009, 95% CI: 7.8%–32.9%). ICU C saw an increase in safety metrics to 90.12%, whereas ICU D improved to 75%. Across various indications, the mean safety rate improved to 84%, underscoring the role of real-time clinical alerts in reducing inappropriate albumin use and enhancing prescribing accuracy. Safety improvement was not only due to better adherence but also unnecessary volume or duration reductions, especially in hypovolemia cases, nonseptic hypoalbuminemia, and ARDS with fluid overload.

An analysis of physician responses to MDSS alerts revealed that 50.9% of prescriptions triggered at least one warning. A total of 133 alerts were generated, with 62.41% originating from ICU C and 37.59% from ICU D [Figure 5]. Heart failure prescriptions exhibited the highest alert rate (15.78%) among specific indications.

Figure 5.

Distribution of issued, responded, and ignored alerts

Importantly, 60.15% of alerted prescriptions were modified following MDSS recommendations, highlighting physician engagement with decision support. The highest rates of alert compliance were observed for conditions such as ARDS, heart failure, and hypovolemia, all of which had a 100% correction rate. In contrast, indications such as paracentesis and pleural effusion exhibited lower response rates, with many alerts being disregarded.

The MDSS exhibited high accuracy, with <5% of alerts deemed false positives. These were predominantly attributed to missing or delayed laboratory data (e.g., albumin levels not yet recorded when prescribing). Physicians or pharmacists typically resolve these cases through manual review. Furthermore, alerts were occasionally overridden without proper documentation, especially in ICU D, underscoring the need for improved user engagement and alert prioritization. Despite these limitations, the system maintained functional stability and usability during the intervention.

All categorical comparisons (e.g., adherence rates and DUE approvals) were assessed using Chi-square tests, whereas continuous variables (e.g., number of vials per prescription) were analyzed using independent-sample t-tests. Significance was set at P < 0.05, and 95% CIs were calculated where applicable. Missing or incomplete data accounted for <2% of total entries and were excluded list wise. No data imputation was performed.

DISCUSSION

This study demonstrated the effectiveness of a MDSS in improving the appropriateness of albumin prescriptions in two ICUs of a tertiary care hospital. The system successfully increased guideline adherence and patient safety, confirming its potential role as an enabler of evidence-based prescribing practices in high-risk clinical settings.

The MDSS significantly improved physicians’ compliance with institutional prescribing guidelines, evidenced by a 20.62% increase in adherence postimplementation. This supports earlier findings that CPOE integrated with decision support can reduce prescribing errors and promote safety.[8,9] The behavior change observed in physician responses to MDSS alerts may parallel the effects of structured educational interventions, such as audits and feedback or point-of-care reminders, which have been shown to improve clinical practice patterns in various settings.[14] In our study, 60.15% of prescriptions with alerts were modified, suggesting a high alert responsiveness rate and clinical trust in the system’s recommendations.

This level of physician responsiveness contrasts with previous literature that reported lower alert acceptance due to alert fatigue, insufficient clinical relevance, or overly rigid algorithm design.[15] The high engagement observed in this study may be attributed to the alignment of alerts with local protocols and integration into daily workflow.

Alert override behavior varied by indication. Conditions such as ARDS, sepsis, and nephrotic syndrome had high correction rates, whereas prescriptions for paracentesis, pleural effusion, and particularly HRS were frequently overridden. In the case of HRS, despite guideline misalignment, physicians often justified albumin prescriptions based on acute decompensation, suggesting a gap between static guideline logic and dynamic clinical judgment.

This finding is consistent with prior work showing that structured CDSS alerts may not fully account for clinical nuance.[16] Addressing this issue may require the integration of adaptive learning algorithms or more flexible, clinician-adjustable thresholds.

Improvement in patient safety – from 63.33% to 82.61% – was not solely due to increased adherence but also reflected optimization of dosage, timing, and duration. As noted by Alshammari,[17] medication safety is a multidimensional construct that encompasses correct indication, route, frequency, and patient-specific risk mitigation. The MDSS contributed to a broader safety culture by proactively preventing inappropriate prescriptions at the point of care. These findings also reflect the strategic importance of digital interventions in minimizing preventable harm – consistent with IOM guidance on preventing medication errors in high-acuity environments.[18]

From a health system perspective, the MDSS streamlined the DUE process. By embedding DUE logic into the prescribing phase, the system reduced administrative burden and enhanced transparency – reinforcing the role of DUE in rational drug use as discussed by Gangwar et al.[5] This aligns with previous calls for real-time, system-level controls beyond retrospective audits.

Moreover, this work contributes to the emerging evidence supporting the role of clinical information systems in emergency and ICU departments, where high variability and urgency elevate the risk of irrational prescribing.[19]

Despite its success, several limitations were identified: The study was conducted in only two ICUs, limiting generalizability across wards or institutions. Physician rationale for alert overrides was not captured formally, limiting interpretation. The system was limited to albumin prescriptions, and performance on other drug classes remains untested. Approximately 40% of alerts were not followed, suggesting residual barriers to trust, usability, or alert relevance.

These findings underscore the need for adaptive, user-centered CDSS design – a point emphasized in pediatric CDSS literature as well.[20] Further effort is needed to reduce false alerts, integrate up-to-date labs, and incorporate flexible override documentation.

Our results are consistent with those of Hashemi et al.,[4] who reported a significant reduction in PICU prescribing errors after CDSS implementation. Similar improvements in alert-driven prescribing were seen in the work of Ibáñez-Garcia et al. and Papandreou et al.[11,13] However, van der Sijs et al. highlighted the high override rates and potential for clinician disengagement when alerts lack specificity,[16] mirroring challenges in our ICU D.

While locally customized, the current MDSS system lacks pharmacogenetic support, which has been increasingly incorporated in advanced CDSS tools like FARMAPRICE – a promising model for future evolution.[21]

To improve scalability and long-term impact, future iterations of this MDSS should consider expansion to multidrug support platforms (e.g., antibiotics, anticoagulants, and high-risk pediatric meds), incorporation of predictive AI models for high-severity cases and individualized risk profiling,[15] addition of cost-effectiveness modules, especially for budget-sensitive agents like albumin, support for pharmacogenomics and dynamic alerting, as demonstrated in,[21] structured capture of override justification to improve algorithm training.

Further research should explore system performance across multiple departments and hospital types, with longitudinal follow-up to assess behavioral sustainability and institutional change.

This pre- and postintervention study demonstrated that integrating a MDSS into ICU prescribing workflows significantly enhanced guideline adherence, improved patient safety, and reduced inappropriate albumin prescriptions in a real-world hospital setting. The system achieved high alert responsiveness among physicians, with over 60% of alerts resulting in prescription changes, particularly in critical conditions such as ARDS and sepsis.

The findings support MDSS’s potential as an effective clinical tool for reinforcing evidence-based practices, minimizing medication errors, and streamlining the DUE process. Importantly, improvements in safety indicators were driven not only by protocol adherence but also by physician engagement with alert content and appropriate dose adjustments.

Nevertheless, the results should be interpreted in light of several limitations, including the single-center, ICU-specific setting and the lack of formal documentation on reasons for alert overrides. The system’s impact on other medication classes and in diverse clinical environments remains to be explored.

To enhance scalability and long-term sustainability, future work should focus on extending MDSS to cover a wider range of drugs, developing adaptive learning models to reduce alert fatigue, incorporating pharmacogenomic data, and conducting economic evaluations to assess cost-effectiveness. In addition, implementation studies in general wards, pediatric units, and community hospitals are warranted to validate such systems’ generalizability and institutional adaptability.

AUTHORS’ CONTRIBUTION

Naghme Dashti: Conceptualization, Data curation, Formal analysis, Methodology, Investigation, Software, Visualization, Writing – review and editing original draft. Hasan Vakili Arki: Conceptualization (lead), Project administration, Supervision, Validation, Funding Acquisition, Writing – review and editing (Lead). Hesamoddin Hosseinjani: Investigation, Project administration, Supervision, Resources, Validation, Writing – review and editing (equal). Saeid Eslami: Conceptualization, Project administration, Validation, Review and editing. Seyed Mohammad Tabatabaei: Methodology, Visualization, Validation, Review and editing.

Conflicts of interest

There are no conflicts of interest.

Funding Statement

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. It was supported solely by the Department of Medical Informatics at Mashhad University of Medical Sciences (MUMS) as part of a PhD candidate's thesis project.