Improving Patient Safety by Modifying Provider Ordering Behavior Using Alerts (CDSS) in CPOE system

Abstract

Medication errors are not unusual in acute care settings. This prospective time series analysis/study evaluates the use of Clinical Decision Support System (CDSS)/alerts in helping providers not to make errors, when putting in orders in a CPOE system. We reviewed electronic health records for all the inpatients coming to 5 community hospitals for a 6 months duration (July 2010 – December 2010). Responses to 9 synchronous alerts (CDSS tools) were studied, that were prompted on computer screens when providers were putting in medication orders in EMR. These alerts guided the providers regarding any drug duplications, interactions, contraindications of the prescribed medicine with patient’s clinical condition etc. The CDSS system in place changed the physician behavior & patient therapy 41.75% of the times when medication orders were placed. These alerts substantially decreased the medication error rate/adverse drug events (ADE’s) in the patients receiving care at these 5 hospitals.

Background

Medication errors are frequent in inpatient settings and the injuries causing them are defined as “Preventable adverse drug events (ADE’s)”. The purpose of having a clinical decision support system (rules/alerts) is to prevent occurrences of as many ADE’s as possible and help the providers to make better and more informed decisions, during ordering process in the EMR. Studies have shown that 2.4–3.6% of hospital admissions are caused by ADE’s of which 69% are preventable.¹ Computerized Provider Order Entry (CPOE) with CDSS has been proven to improve patient safety. ^2–4 Most of the benefits of the CPOE system result from integrated clinical decision support tools for drug overdoses, interactions, allergies etc. ⁵ These tools provide real time alerting & warning to providers (MD’s, DO’s, NP’s, PA’s, CRNA’s), pharmacists & RN’s taking verbal orders from a provider, for any such unfavorable outcomes, when they are putting in orders for patient care in EMR’s. Studies at various institutions noted 69–80% of alerts were overridden by physicians in acute care settings. ^6,7 Our objective in this study was to evaluate and analyze the change of provider behavior & its eventual effect on patient therapy, when putting in orders using the CPOE system, with Clinical Decision Support System (Synchronous alerts) in it. In the CPOE application analyzed here, these alerts could be accepted or overridden. The system in place in these 5 hospitals allowed 4 ways to address these alerts. Based on the way the provider addressed these alerts, the alerts were identified and classified as the ones which change provider behavior and the one which did not (Table 1).

Table 1:

Characteristics of addressing alerts with examples and understanding its effect on Provider behavior

Alert Resolution & its impact on Ordering behavior	Examples	Changes Provider behavior
Cancelled alert & alerting order with action taken	An alert comes when the provider is entering an order for an anticoagulation medication and the patient does not have a recent PT/PTT or INR lab result within a specified number of days. The provider in this case cancels the anticoagulation order and takes action of ordering a PT/INR on this patient.	Yes
Cancelled alert & alerting order with no action taken	An alert comes when the provider is entering an order for an anticoagulation medication and the patient does not have a recent PT/PTT or INR lab result within a specified number of days. The provider in this case cancels the anticoagulation order and takes no further action at that point of time.	Yes
Overrided alert & did not cancel the order, but relevant action taken	An alert comes when the provider is entering an order for an anticoagulation medication and the patient does not have a recent PT/PTT or INR lab result within a specified number of days. The provider in this case overrides the anticoagulation alert, does not cancel the order but takes action of ordering a PT/INR on this patient.	Yes
Overrided alert, did not cancel order and no action taken	An alert comes when the provider is entering an order for an anticoagulation medication and the patient does not have a recent PT/PTT or INR lab result within a specified number of days. The provider in this case overrides the anticoagulation alert, does not change the alerting order and takes no further action at this point of time.	Yes

Methods

Study Sites & System details

Electronic Medical Records were evaluated for a 6 month duration (July 2010 – December 2010) for all the patients who received inpatient medical/surgical care in 5 multi specialty community hospitals (part of Adventist Health System) having total of 514 acute care beds across 3 states (Table 2), which were on CPOE system (Cerner). The hospitals chosen for this study were Takoma Regional Hospital (TN), Florida Hospital Zephyrhills (FL), Emory Adventist Hospital (GA), Gordon Hospital (GA), and Park Ridge Hospital (NC). The sites chosen were on the same electronic platform, using the same version/platform of vendor and same content for CPOE & CDSS. The details of these hospitals are provided in Table 2.

Table 2:

Details of hospitals that were chosen to evaluate the alert responses

Hospital name	Number of Acute care beds	Number of Providers	Date of CPOE Go-live
Takoma Regional Hospital (Pilot site)	100	120	5/12/2009
Florida Hospital Zephyrhills (Pilot site)	154	160	6/9/2009
Gordon Hospital	69	65	4/20/2010
Emory Adventist Hospital	88	240	5/4/2010
Park Ridge Health	103	230	5/18/2010

Classification of Alerts

The EMR system studied in these hospitals had both Asynchronous and Synchronous alerts. Our study only analyzed 9 synchronous alerts in the system which prompted the providers/end users on the computer screen on which they were working (Table 3). These synchronous alerts were prompted to end users/providers when ever the clinical decision support tools prepared in the computer system identified the defined orders and related clinical conditions programmed in it. These rules were chosen & defined on most relevant & frequently observed clinical conditions which might have interactions to orders (that patients receive during their inpatient stay in the hospital) and have the capacity of creating ADE’s & negative patient care safety outcomes. The EMR had many hundred more asynchronous alerts, which did not prompt the providers/end users on the computer screens, but printed instructions/warnings to other care givers at the back end. An example of one such asynchronous alert could be a print out to pharmacy when the INR of patient is more then 4.

Table 3:

Classification of alert types with the reasoning of firing of the alerts

Alert Type/Nomenclature	Explanation of the reasoning for firing of this alert
Increased Bleeding Alert	Enhances the care provider’s awareness of situations of increased risk for bleeding related to adverse events. Patients may be at an increased risk of bleeding when invasive routes of administration such as Intramuscular, Intrathecal, etc. are combined with anticoagulant medications like heparin or low molecular weight heparins (LMWH). This rule compares the current scratchpad order with existing orders against the medication and the route of administration to determine whether or not to provide an alert to the user. Routes evaluated are: intrathecal, intraspinal, IM, and Epidural.
Sleep Apnea Alert	Present an alert to providers ordering narcotics and other medications that depress respiration when the patient does not have an existing order for CPAP has a current health history (Sleep Apnea Medical Hx or Current Home Treatment) documentation of Apnea or CPAP use Presents the Pharmacists with an alert that does not have the RT CPAP order option and prompts them to contact the ordering physician to evaluate the need for CPAP.
Radio Contrast Media Alert	Patients with impaired renal function and dehydration are at risk of developing radio contrast media-induced nephropathy. The rule also looks for any Metformin orders in ordered status or discontinued within the last 24 hours. If a patient has a new result of with a platelet count less than 100 or is now at a 50% decrease over the last seven days and has an active order for a drug which can cause thrombocytopenia, an alert is printed in the pharmacy. This rule also looks for an active order for chemotherapy meds and does not fire off if one is active.
Digoxin Toxicity Alert	Warn potential digoxin toxicity when ordering digoxin in patient with recent laboratory values which would predispose patient to developing an ADE. Alert evokes when digoxin is ordered. Recent laboratory values are checked for low potassium, or magnesium blood level, or an already high digoxin level. Recent orders are also checked for current potassium, or magnesium, supplementation which would indicate the electrolyte problem is already being addressed Alert evokes when ACE inhibitor, potassium supplement, potassium-sparing diuretic is ordered. Recent laboratory values checked for high potassium level and orders checked for current supplementation.
Potassium Toxicity Alert	Warn of potential toxicity when ordering a drug which may exacerbate an existing high serum potassium level Alert evokes when heparin order is added to the scratchpad. Recent PTT values are then checked for prolonged bleeding times which may be indicative of over coagulation.
Anticoagulation Alert	To provide a warning of potential coagulation compromise when ordering heparin on a patient with recent PTT values that would predispose the patient to developing an adverse drug event. The medication lists are pulled from the Creatinine Clearance asynchronous rules.
No Creatinine Alert	This rule will present an alert when the provider is entering an order for a nephrotoxic or renally excreted medication and the patient does not have a recent Creatinine Serum lab result within a specified number of days.
No PT/PTT/INR Alert	This rule will present an alert when the provider is entering an order for an anticoagulation medication and the patient does not have a recent PT/PTT or INR lab result within a specified number of days. The medication lists are pulled from the Anticoagulation asynchronous rules.
Pregnancy/Lactation Alert	This rule will present an onscreen alert when a medication that has been identified as a risk is ordered and the patient is pregnant or lactating.

Clinical analysis of alerts

Orders can be entered in the EMR by a provider (any licensed practitioner), pharmacist or Nurses. The corporate policy in place in the Adventist Health System allowed nurses to take verbal orders on the phone from providers, but only with providers holding the phone until the nurse puts in the verbal orders in computer and reads the order back to provider, once he/she completes the order in the EMR. This way of verbal orders provided safe and effective means of medication orders being placed in the system and prevented a situation where nurses when prompted with alerts had to adress them on their own; as the providers are on the phone, they address these alerts by informing the nurse on the phone about appropriate response to these prompts, if they ever come during the verbal order process. We generated the data of all such alerts from Cerner EMR for orders entered in 6 months period, from January 2010 to July 2010. All the 5 hospitals had CPOE in place during this period of time, although for different durations of time. The synchronous alerts identified and analyzed in our system were triggered by a certain set of orders and they prevented occurrences of Adverse Drug Events in each category (Table 4)

Table 4:

Analysis of the various orders generating the alerts and potential ADE’s caused by situations addressed by alerts

Alert Type/Nomenclature	Common Orders generating this alert	Potential ADE’s Prevented
Increased Bleeding Alert	Heparin, Warfarin, Enoxaprin, etc.	Bleeding
Sleep Apnea Alert	Hydrocodone, Acetaminophen-oxycodon, Acetaminophen-hydrocodone, Acetaminophen-propoxyphene, Butarphanol, Alprazolam, Chlordiazepoxide, Clonazepam, Codeine, Diazepam, Droperidol, Etomidate, Fentanyl, Meperidine, Methadone, Methohexital, Midazolam, Morphine, Nalbuphine, Opium, Oxazepam, Oxycodone, Phenobarbital, Propoxyphene, Propofol, Temazepam, etc.	Respiratory Depression
Radio Contrast Media Alert	CT Scan with or without contrast, CT Angiograms, MR with or without contrast, MR Angiogram, Fluoroscopic guided special procedures, Interventional radiology procedures using dyes, Intravenous Pyelographies, etc.	Anaphylaxis
Digoxin Toxicity Alert	Digoxin	Electrolyte Imbalance, Hypokalemia
Potassium Toxicity Alert	Enalpril, Lisinopril, Quinapril, Amino acids, Spironolactone, Potassium Chloride, Potassium Phosphate, TPN Orders	Hyperkalemia
Anticoagulation Alert	Heparin	Bleeding
No Creatinine Alert	Amikacin, Gentamicin, Glyburide-metformin, Metformin, Neomycin, Tobramycin, Vancomycin, etc	Nephrotoxicity
No PT/PTT/INR Alert	Argatroban, Cephalaxin, Dalteparin, Enoxaparin, Fondaparinux, Heparin, Lepirudin, TPN Orders, Warfarin	Bleeding
Pregnancy/Lactation Alert	Aminoglycoside Antibiotics, Antipsychotics, Antianxiety, Aspirin, Amlodipine, Statins, Azathioprine, Bethenacol, calcitriol, Diclofenac, Dicycolmine, Doxazosine, Ephedrine, Hydroxychloroquine, Imipramine, Estrogens/Progesterone’s, Methimazole, Methotrexate, Paroxitine, Penicillamine, Azoles, Warfarin, etc	Secretion of toxic agent in lactating mother causing toxicity to newborn/lactating child

Main Outcome of the Measure & Results

All of our responses (Table 5) in terms of the number of alerts addressed and how the providers responded to these alerts had a pattern. Total 9705 synchronous alerts were evaluated for this duration of time, out of which 28 alerts (0.29%) caused the cancellation of potentially harmful order, with action taken by provider; 995 alerts (10.25%) caused potentially unsafe orders cancelled with no action taken by provider; 3011 alerts (31.03%) were overridden, alerting order unchanged, but action taken by providers: & 5671 alerts (58.43%) were overridden, alerting orders unchanged with no action taken by the provider. Out of the 4 kinds of interventions for each alert explained above, the first 3 interventions changed the provider ordering behavior, by either cancelling the order (taking or not taking any action after the cancellation of the order) or overriding the alert with an action taken for that order after being overridden. These 3 interventions aggregated to 41.57% of total times the orders were placed. A few False Positive alerts, which were generated by Data base administrators (Non Providers) on test patients, were not included in the statistical analysis done above.

Table 5:

Breakdown of the Alert types with provider response (Cancelled/overridden) to the alerts and numerical analysis of the final results.

Rule Description	Alert & Order Cancelled with Action	Alert & Order Cancelled without Action	Alert Overridden, Order unchanged, with Action taken	Alert Overridden, Order unchanged without Action taken	Total False Positive	Final Total
Increased Bleeding Alert	0	20	0	26	1	46
Sleep Apnea Alert	0	134	298	1424	10	1856
Radio Contrast Media Alert	0	266	35	1297	28	1598
Digoxin Toxicity Alert	0	3	0	26	0	29
Potassium Toxicity Alert	0	9	0	8	1	17
Anticoagulation Alert	0	7	0	20	0	27
No Creatinine Alert	1	152	206	706	33	1065
No PT/PTT/INR Alert	27	302	2472	1350	86	4151
Pregnancy/Lactation Alert	0	102	0	814	2	916
Grand Total	28	995	3011	5671	161	9705
Percentage Behavior Change	0.29	10.25	31.03	58.43		100.00
Changes Provider Behavior & Patient Therapy	Yes	Yes	Yes	No

Discussion

Several studies have been done in the past that have outlined the benefits of the CDSS and especially with CPOE. Rules and alerts that come with electronic order entry have been analyzed in a study similar to ours at VA health system in the CPRS System. ⁸ Their study analyzed the severity of order checks, including Drug-Drug duplication & Drug-Allergy order checks for duration of 5 years. Our study is more about understanding the clinical alerts, how these alerts are generated in different clinical conditions and analyzing the change in the provider behavior due to these alerts in a 6 month period of time. In some studies Nurses and NP’s added to the pool of providers addressing the alerts. ^9–12 Our pool of providers included Nurses along with all the licencesed practitioners (MD’s, DO’s, NP’s, PA’s, CRNA’s), yet in our settings the policy of read back of verbal orders to the providers on the phone made the order entry process safer then other institutions.

Conclusions

We conclude from this prospective time series study that the Clinical Decision Support System (CDSS) in our EMR provided safer care to the patients coming to these 5 hospitals. Monitoring of CDSS in place changed the provider ordering behavior by 41.75% of the time alerts were raised. This study contributed to improved understanding of the alert responses & helped in prevention of unintended consequences of drug interactions/duplications/allergies due to lack of provider information of patient condition if these orders were written on paper. This change in ordering behavior in providers substantially lowered occurrence of ADE’s in these patients, including conditions like bleeding, respiratory depression, anaphylaxis, electrolyte imbalances (Hyperkalemia, Hypokalemia), nephrotoxicity & secretion of toxic agents in lactating mothers. Goal of our study was to analyze the percentage improvement in provider ordering behavior using clinical decision support system in our EMR & indeed the results of our study demonstrated that providing most legible and relevant prompts to our providers delivers the safest patient care in our hospitals.

Figure 1:

Medication order (Unasyn) generating a synchronous alert showing a drug duplication. A provider has option of cancelling the order or overriding the alert