Keywords
clinical decision support - adverse drug event - safety - pharmacy - alerting
Background and Significance
Background and Significance
Medication errors significantly increase patient morbidity and mortality and cause
considerable increases in costs in health care institutions.[1] Most are considered to be preventable.[2]
[3]
Medication errors can occur at any point in the medication use process. However, research
has shown that they are more common at the prescription stage (39%).[4] Different approaches have been adopted to help physicians minimize these errors,
including educational strategies and use of advanced health information technologies.
Major emphasis has been placed on implementing computerized physician order entry
(CPOE) combined with a clinical decision support system (CDSS).[5]
[6]
[7]
[8]
[9] CDSS provides clinicians with clinical knowledge and patient-specific information
in order to optimize the safety and quality of the pharmacological treatment prescribed.
These systems are associated with reduced morbidity rates, improved prescribing practices,
improved patient monitoring, reduced health care costs, and reduced adverse drug event
(ADE) rates.[10]
[11]
[12]
[13]
Optimal use of CPOE with CDSS requires integration of multiple clinical information
systems, including medical records, clinical laboratory software, and pharmacy-based
software. This is not a concern for hospitals where the electronic health record already
integrates clinical data. However, in hospitals such as ours, where several clinical
information systems are being used, data integration has proven to be challenging.
During the development and implementation of the CDSSs, other numerous barriers are
encountered. The majority of CDSSs are not able to integrate the information systems
in real time or to translate clinical guidelines into appropriate alerts.[14]
[15] The lack of sensitivity and specificity of the alerts, together with the inability
to customize them, often results in a high rate of override.[16]
In 2015, our institution developed a CDSS (HIGEA) that benefits from the integration
of multiple hospital information systems in real time and generates patient-specific
alerts to prevent potential ADEs based on predefined consensual clinical rules (CRs).
Alerts are reviewed by the pharmacists during their ward-based activities in order
to prevent alert fatigue during prescription. CRs can be redesigned and customized
in order to improve their effectiveness.
Objectives
The objective of the present study was to evaluate the effectiveness of this automated/integrated
real-time CDSS in the prevention of potential ADEs.
Methods
Setting
Our hospital is a 1,300-bed tertiary teaching institution employing more than 8,000
professionals who are responsible for the direct specialist health care of a catchment
population of approximately 350,000 people. [Table 1] shows the most significant indicators of hospital care activity between January
2016 and June 2016.
Table 1
Indicators of hospital care activity between January 2016 and June 2016
|
Indicator
|
N
|
|
Hospitalizations
|
25,449
|
|
Hospital stays
|
190,799
|
|
Average length of stay
|
7.5 d
|
|
Surgical procedures
|
16,772
|
|
Medical consultations
|
329,487
|
|
Drug prescriptions/day
|
11,500
|
In our hospital, physicians enter all the prescriptions into an in-house CPOE system,
which was traditionally supported by a basic CDSS (Farhos Prescriplan – Visual Limes,
Spain). This safety monitoring tool is based on protocols, standardized doses for
most drugs, and alerts in case of allergies, duplications, and interactions. The process
is automated using a real-time alert system. When a medication is prescribed, an alert
generated in the system warns the doctor of any possible medication error. Immediately
after prescription, the CPOE system is checked for new or modified medication orders
placed by the pharmacist. The pharmacist ensures continuous centralized order verification.
Twelve pharmacists are responsible for patient care areas during the day shift; each
pharmacist covers one specialty area. Dispensing is facilitated by profiled automated
dispensing cabinets (Pyxis, Grifols), from which nurses can withdraw medication once
it has been prescribed and verified by the clinical pharmacist.
Design of the Advanced CDSS: HIGEA
HIGEA was designed by a multidisciplinary team comprising 10 permanent members. The
pharmacy department was represented by four pharmacists from different areas of expertise
(medical, surgical, critical care, and a dedicated ADE pharmacist) and the medical
staff by a nephrologist, a hepatologist, a hematologist, and an infectious disease
specialist. The team was completed by two information systems specialists. Team members
were selected based on their extensive clinical experience. In any case, all members
were asked to present their colleagues' opinions and to consult them if necessary
during the development of the system.
HIGEA integrates multiple hospital information systems in real time and generates
patient-specific alerts to prevent ADEs based on predefined CRs ([Fig. 1]).
Fig. 1 Design of clinical decision support system.
The integration was designed to receive Health Level Seven (HL7) messages from laboratory
systems and the CPOE. A Mirth server was setup, and several channels were set to work
in real time in order to receive the information sent from these systems. A few seconds
after a message is received, a Web service is available to provide information regarding
existing alerts for the patient in the message. Therefore, if a physician prescribes
a new drug or new laboratory information is generated, HIGEA automatically activates
or deactivates the alert. An average volume of 50,000 messages are processed daily.
The information was transformed to a “standard” data structure in a NoSQL database,
where all information of a given patient is stored as a document. This data store
format enables us to check very quickly whether a given patient satisfies a specific
rule or not, as complete patient information is retrieved at once. Unlike other, similar
solutions, which are based on traditional SQL systems, HIGEA does not need to perform
SQL joins to find information on a given patient, thus saving processing time and
improving real-time responses.
The system was developed with the following features:
-
Integration of clinical laboratory data and the CPOE system. Laboratory values (biochemistry,
hematology, immunology, and genetics) were imported from the laboratory online results
program, Modulab. Data on patient characteristics and drug use were imported from
the CPOE, Farhos Prescriplan. The integration was performed using a standard language
(HL7) and algorithms that create a homogeneous knowledge base to process CRs.
-
Generation of alerts by combining data from the electronic patient databases (laboratory
data and CPOE) and the CR bundle previously defined by the multidisciplinary team.
The detection system ran in real time and performed searches for new orders or new
laboratory test values.
-
Generation of standardized recommendations for the pharmacists' interventions. The
multidisciplinary team agreed on the advice that the pharmacist had to give to prescribers.
-
Automatic recording of the pharmacists' interventions and their acceptance by the
physicians. The interventions that pharmacists implemented to prevent a potential
ADE in response to the alerts generated by HIGEA were automatically recorded by the
Web service in real time. This collects and stores all the information received by
the information systems. Thus, the pharmacist only reviews the information recorded
by the system.
-
Generation of dashboards that allow analysis of data and a systematic evaluation of
the usefulness of the system.
-
Automatic prioritization of alerts based on their impact. In order to prevent alert
fatigue, the rules were continually assessed by evaluating its positive predictive
value (PPV). Rules that have more impact on patients are highlighted and shown first
in the Web service. The pharmacist can then review the most relevant alerts first.
Definition of Clinical Rules: A Consensus-Based Process
A CR consists of the observation of ≥ 1 clinical positive and/or negative condition
(e.g., a drug prescription and the presence of a laboratory value) and a standardized
recommendation to change the treatment in order to prevent an ADE. A CR bundle was
defined for four different intervention programs: (1) program for dose adjustment
in renal impairment; (2) program for adjustment of anticoagulation/antiplatelet therapy;
(3) program for detection of biochemical/hematologic toxicities; and (4) therapeutic
drug monitoring program (detecting inappropriate blood levels of high-risk drugs).
A dedicated ADE pharmacist performed a bibliographic search to identify possible CRs
for each of the four intervention programs defined. These CRs were presented to the
multidisciplinary team in order to identify those that, according to clinical practice,
could be more effective in the prevention of ADEs. The team prioritized those CRs
that were considered more harmful, less known by physicians, less prevalent, or more
complex. The final definition of the bundles required 8 meetings, 2 for each program,
with each lasting approximately 3 hours. In addition, a 2-hour introductory session
was held to explain how the CDSS operated.
Software Validation
After building the CRs into HIGEA, a validation process was carried out over 3 months
to establish the technical correctness of the system. The pharmacists reviewed the
alerts reported by HIGEA and, after checking the patient's clinical information, ensured
that the system adequately detected patients described in the expected situations
in the CR. They then contacted the prescribing physician to recommend changes in treatment.
This process ensured that the system does not generate false-positive or false-negative
alerts. Once it was validated that the software fully satisfied all expected requirements,
it was implemented in daily clinical practice.
Validation Study
In order to assess the true impact of alerts generated by the CRs, we performed an
observational nonrandomized prospective study from January 1, 2016 to June 30, 2016.
The PPV was analyzed as an indicator of the effectiveness of the previously defined
CRs. The PPV is the probability that an alert can prevent a potential ADE, namely
as the ratio of modifications in treatment to alerts reviewed. It measures the real
impact of a CR on a patient, which is theoretical.
PPV = alerts with an accepted intervention / total of alerts.
The total number of alerts generated by the system, the prescribing errors intercepted,
and the response to the intervention were analyzed. The specific intervention program
that enabled the detection of the potential ADE and the patient's location by department
was also identified.
This study was conducted in accordance with the Declaration of Helsinki and was approved
by the ethics committee of the Gregorio Marañón Hospital before the study began.
Results
The team defined 211 CRs: 110 for the “program for dose adjustment in renal impairment,”
24 for the “program for adjustment of anticoagulation/antiplatelet therapy,” 64 for
the “program for detection of biochemical/hematologic toxicities,” and 13 for the
“therapeutic drug monitoring program.” Based on these CRs, HIGEA generated a total
of 1,086 alerts (8.9 alerts per working day) during the study period. The impact of
the alerts in terms of the number of interventions generated and the grade of adherence
to the recommendations are shown in [Fig. 2].
Fig. 2 Number of alerts and rate of intervention.
The pharmacist reviewed all the alerts and approximately half of the alerts (554 out
of 1,086 [51%]) generated an intervention to prevent a possible ADE. Of these, 368
interventions were accepted by health care professionals and led to a change in treatment
(e.g., changes in dose or frequency or route of administration, end of treatment).
Overall, 368 of 1,086 alerts (34%) required a documented modification to therapy because
of a real prescription error intercepted. No advice was given in 532 of 1,086 alerts
(49%). In these cases, the pharmacist evaluated the patient's clinical status by consulting
the clinical history and contacting the prescribing physician. Then, the prescribed
treatment was considered adequate (e.g., furosemide and sodium [Na] < 125 mmol/L:
furosemide can induce hyponatremia but is indicated in patients with hypoosmolar hyponatremia;
levofloxacin and estimated glomerular filtration rate [eGFR] < 50 mL/min: a patient
with eGFR = 30 mL/min and levofloxacin 250 mg/12 h is correct for a patient with complicated
pneumonia but not for a patient with bronchitis).
[Table 2] shows all the results for alerts generated by intervention program and the global
PPV in each intervention program. The CRs with the highest PPV in each intervention
program are detailed in [Supplementary Table S1] (available in the online version).
Table 2
Positive predictive value of alerts generated by intervention program
|
Outcome
|
Alerts
(N)
|
Alerts with intervention
(N)
|
Treatment changes induced
(N)
|
Positive predictive value
|
|
Intervention program
|
|
|
|
|
|
Program for adjustment of dose in renal impairment
|
430
|
306
|
218
|
0.51
|
|
Program for adjustment of anticoagulation/antiplatelet therapy
|
395
|
145
|
94
|
0.24
|
|
Program for detection of biochemical/hematologic toxicities
|
212
|
93
|
46
|
0.22
|
|
Therapeutic drug monitoring program
|
49
|
10
|
10
|
0.20
|
|
Total
|
1,086
|
554
|
368
|
|
[Fig. 3] shows the analysis of alerts generated in each program stratified by type of unit.
Fig. 3 Acceptance rate in each intervention program stratified by type of unit.
Overall, the highest number of alerts with an intervention (396; 71%) was generated
for patients admitted to medical units, followed by surgical units (88; 16%) and critical
care units (70; 13%). The percentage of accepted interventions was similar in surgical
units (68%), medical units (67%), and critical care units (63%).
The most frequent individual alerts that led to changes in treatment were generated
in response to enoxaparin subcutaneous and eGFR < 30 mL/min (55/368 changes), ranitidine
orally and eGFR < 50 mL/min (39/368 changes), meropenem and eGFR < 50 mL/min (26/368
changes), omeprazole and Na < 125 mmol/L (20/368 changes), and levofloxacin intravenous
and eGFR < 50 mL/min (18/368 changes).
Discussion
We describe the implementation of an advanced CDSS and evaluate its utility in identifying
potential ADEs. Specifically, we confirm that real-time integration of clinical information
provides a highly efficient CDSS for improving medication safety. When specialist
pharmacists examine the alerts generated daily, the system identifies a large number
of potential ADEs that go unnoticed by the physician and led to a change in the clinician's
decision (368/554; 66%). These highly effective alerts are prepared to be incorporated
into the CPOE.
The major benefit of this solution is that it combines data from various clinical
information systems (clinical laboratory data and CPOE system), whereas most other
CDSS are severely limited by nonintegrated information systems[17]
[18]
[19]
[20] and, therefore, less specific rules that are restricted to a single criterion such
as prescription of a medication or an abnormal laboratory result. These rules have
high false-positive rates because their occurrence is rarely related to an ADE. The
PPV of these rules is < 0.1, which is much lower than our values (0.34). Recently
published studies on medication-related CDSS identify the increase in the sensitivity
and specificity of alerts as an area that can be improved. This improvement should
include more patient-specific information in order to reduce alert fatigue.[21]
[22]
Second, alerts run in real time, and the system continuously performs searches for
new orders and laboratory test values. This feature differs from many commercially
available CDSS, in which real-time analysis is not available. Continuous searching
also helps to obtain more efficient CRs with a better PPV, because the alerts always
show the real clinical situation of the patient (the most recent drug prescription
and laboratory test value).
It is also noteworthy that HIGEA uses CRs to detect patients at risk of an ADE instead
of using alerts to identify ADEs that have already occurred. This approach differs
from other software applications in which rules commonly include a toxic serum drug
level or prescription of an antidote.[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30] In fact, Silverman et al[30] demonstrated that after modification of their CRs from detecting actual ADEs to
identifying potential ADEs, the volume of interventions by the pharmacists increased
and the rules became more efficient.
HIGEA was designed as a customizable system, which allows the user to modify the CRs
in order to increase their efficiency. Other similar systems described by the scientific
literature do not include this facility, and the incorporated commercial bundle of
CRs cannot be modified by the user.[30]
[31]
Finally, another major benefit of HIGEA is that it was provided with a large set of
CRs that covers a wide range of clinical conditions and provides a specific dosage
recommendation for the new prescription.[32] Only 10% of published studies in this area[15] use rules that have been defined according to clinical needs. In our case, the CRs
were defined by a multidisciplinary team that applied a structured method for the
various intervention programs. It has already been demonstrated that an updated and
consistent knowledge base positively affects the efficiency of these systems.[31] The “program of dose adjustment in renal impairment” was prioritized because of
the high number of drugs that have to be adjusted in patients with renal impairment
and the very scant knowledge of these drugs among health professionals. In fact, it
has been demonstrated that the rate of appropriate drug prescribing in kidney impairment
is low and that the use of a CDSS could improve patient safety.[32]
[33] The “program for adjustment of anticoagulation/antiplatelet therapy” was prioritized
because of the severe events that an error with such high-risk drugs could lead to.
The “program for detection of biochemical/hematologic toxicities” was included because
most of these ADEs are not often taken into account owing to their low prevalence.
The “therapeutic drug monitoring program” was included because the alert system can
shorten the time to response once the drug level is available.
Although the percentage of safety alerts resulting in an intervention by the hospital
pharmacist (51%) may seem low, it is significantly higher than the percentages observed
with other CDSS.[15]
[34]
[35] Nanji et al[36] evaluated the rate of alert overrides (alerts without an intervention) and its appropriateness.
Surprisingly, and consistent with our results, they found that about half of the CDSS
alerts were overridden by providers and consequently did not result in an intervention.
The highest number of alerts with an intervention was generated for patients admitted
to medical units (71%). Given the considerable effort necessary to implement a new
technology and in view of these results, this type of patient would be prioritized.
Two-thirds (66%) of the interventions were accepted by physicians; this finding is
consistent with the percentages observed in other studies. In our case, we did not
identify significant differences between medical, surgical, and critical care units
(63–68%). Jha et al[34] investigated a commercial computerized surveillance system (Vigilanz Corporation
– Dynamic Pharmacovigilance) over a 4-month period and found that the physician was
contacted in 30 of 266 reviewed alerts (11.3%). The acceptance rate of these 30 interventions
was 50%. Kilbridge and Ahmad[35] evaluated 4,604 triggers from a computer-based ADE surveillance system over a 2-month
period; of these, 260 led to an intervention (4%). After analyzing the Theradoc computer-based
monitoring system, Silverman et al[30] reported an intervention rate of 5 to 13% and an acceptance rate of 78 to 92%. Rommers
et al[37] evaluated the use of a CDSS (“ADEAS”) similar to ours over a period of 5 months.
In this case, the system generated 2,650 alerts, of which 204 led to an intervention
(7.7%). The percentage of acceptance was 63%.
It should be noted that the acceptance percentages are low, probably owing to the
complexity of the clinical situation of some patients, which sometimes forces us not
to follow standard recommendations in daily clinical practice.
Finally, it is noteworthy that none of these studies were performed in Europe.[15] Despite growing evidence of the positive clinical impact of health technologies
on safety, their adoption and implementation are very slow in some countries. In Spain,
according to a survey conducted by the Spanish Society of Hospital Pharmacy in 2015,
only 20% of hospitals had implemented a CDSS aimed at increasing the safety of pharmacotherapy.[38] High cost is a major barrier, since hospitals must make a large initial investment
with no clear return. For these reasons, there is an urgent need to test CDSS in Europe
because of significant structural differences in health systems between both regions.
We consider that our study provides relevant results for making strategic decisions
concerning implementation of measures to increase patient safety.
HIGEA has been designed in such a way that it can be exported to other institutions;
both the software and the CRs were validated in this study. In fact, three Spanish
hospitals are already using it.
Limitations
First, not all potential ADEs can be detected with this technology, as some are related
to clinical information expressed using natural language, as is the case with diagnosis
and symptoms. This information is not easily accessible, although it has already proven
to be very useful for other authors.[39]
[40] Since HIGEA does not fully integrate the electronic health record, the detection
of such ADEs is not possible and, consequently, some CRs present a low PPV. We are
currently working to improve reasoning by including natural language processing and
identification of semantic entities from unstructured information in the electronic
health record.
Second, HIGEA was developed as a CDSS within the CPOE. However, to evaluate its effectiveness
in the prevention of ADE it is necessary guarantee the analysis of all the generated
alerts, which will allow us to reliably calculate the PPV for each of the CRs. For
this reason, the alerts generated by the system are reviewed by the pharmacists and
not directly by the prescribing physician. We believe that prescreening of alerts
by a clinical pharmacist reduce alert fatigue during prescription, and could increase
the likelihood of appropriate prescribing of these medications.[10]
[31]
[41]
[42] In fact, a review from van der Sijs et al[16] showed that physicians ignore safety alerts in 49 to 96% of cases. In the future,
some alerts might be better presented as online alerts for the physician when immediate
action is necessary, that is, those with PPV = 1.
Finally, given that the recommendations defined in the program for adjustment of dose
in renal impairment do not include specific indications for obese or hemodialysis
patients, these groups were excluded from the study.
Conclusion
HIGEA, an integrated real-time CDSS, is highly effective in preventing potential ADEs.
The clinical pharmacist has played a key role in the success of the system. Our study
offers evidence that customization of CRs significantly improves the safety and quality
of health care decisions when a multidisciplinary team is involved.
Clinical Relevance Statement
Clinical Relevance Statement
HIGEA is a CDSS that makes it easier for health care professionals to identify patients
with a high risk of experiencing an ADE, thanks to real-time integration of various
hospital clinical information systems.
Multiple Choice Questions
Multiple Choice Questions
-
Which of the following CDSS is more efficient in the detection of potential ADEs?
-
A CDSS that receives the drug prescription information.
-
A CDSS that receives the drug prescription information and clinical laboratory data.
-
A CDSS that receives the drug prescription information, clinical laboratory data,
and electronic health record information.
-
A CDSS that receives clinical laboratory data.
Correct Answer: The correct answer is option c, because the system combines data from many hospital
information systems, thus enabling the use of clinical rules with a higher positive
predictive value.
-
What is one of the major benefits of an advanced clinical decision support system?
-
It provides clinicians with clinical knowledge.
-
It enables drug prescription by physicians.
-
It generates alerts by combining data from the electronic patient database.
-
a and c are correct.
Correct Answer: The correct answer is option d, because these are the two functionalities that differentiate
our system from a simple clinical decision support system.
