ENABLING DRUG-DRUG INTERACTIONS IN AN ELECTRONIC MEDICATION - - PowerPoint PPT Presentation

enabling drug drug interactions
SMART_READER_LITE
LIVE PREVIEW

ENABLING DRUG-DRUG INTERACTIONS IN AN ELECTRONIC MEDICATION - - PowerPoint PPT Presentation

ENABLING DRUG-DRUG INTERACTIONS IN AN ELECTRONIC MEDICATION MANAGEMENT SYSTEM: IMPACT ON PRESCRIBER ALERT BURDEN Anmol Sandhu BPharm (Hons), MRes (Health Informatics) Health Data Analytics Conference, October 2019 Electronic medication


slide-1
SLIDE 1

ENABLING DRUG-DRUG INTERACTIONS IN AN ELECTRONIC MEDICATION MANAGEMENT SYSTEM:

Anmol Sandhu – BPharm (Hons), MRes (Health Informatics) Health Data Analytics Conference, October 2019

IMPACT ON PRESCRIBER ALERT BURDEN

slide-2
SLIDE 2

Electronic medication management (EMM) systems

HEALTH DATA ANALYTICS - OCT 2019

Prescriber’s medication

  • rder

Pharmacist’s review of medication

  • rder & supply
  • f medicine

Nurse’s documentation administration

  • f the medicine

And all the processes in between…

  • CPOE or ePS
  • Electronically support the medication management process
slide-3
SLIDE 3

Clinical Decision Support (CDS)

  • Clinical decision support is an intelligent feature commonly integrated within EMM

systems

  • Reduction in medication errors1,2
  • Provides relevant clinical content and patient data to:
  • facilitate clinical decision making
  • notify of a potential adverse outcome
  • Alerts that trigger at the point of prescribing are a common form of CDS
  • Alerts can target different clinicians – nurses, pharmacists, doctors

HEALTH DATA ANALYTICS - OCT 2019

slide-4
SLIDE 4

Drug-drug interaction Therapeutic Duplication Dose Range breach Drug-allergy interaction Local Restriction rules

Types of CDS alerts

HEALTH DATA ANALYTICS - OCT 2019

Prescribers can ‘accept’ or ‘override’ (bypass) alerts

slide-5
SLIDE 5

Prescriber alert burden

Alert volume experienced by prescriber

Alert fatigue

HEALTH DATA ANALYTICS - OCT 2019

WHAT IS IT? HOW CAN IT BE MEASURED?

HIGH ALERT BURDEN

  • Rate of alerts encountered per prescriber
  • Proportion of a prescriber’s medication orders that generate alert
  • Prescriber-centric alert burden data lacking

Ignore all alerts, including critical

  • nes

User frustration High override rates Threshold for alert fatigue is unknown

slide-6
SLIDE 6

St Vincent’s Hospital, Sydney (SVHS)

HEALTH DATA ANALYTICS - OCT 2019

STUDY SITE

  • 379 bed tertiary referral hospital
  • Complex specialties including Heart, Lung,

Bone Marrow Transplantation, AIDS/HIV, Cardiology, Cancer care

  • EMM system implemented in 2005

(MedChart – DXC Technology)

  • Used in all inpatient wards, including ICU
  • My role at SVHS: Specialist Electronic

Medicines Management Pharmacist

slide-7
SLIDE 7

CDS at St Vincent’s Hospital

HEALTH DATA ANALYTICS - OCT 2019

Drug-Allergy Therapeutic Duplication Dose Range (limited) Local restriction rules (limited)

Passive (order sentence, sets, on-demand look up)

Drug-drug interaction (DDI) alerts

  • Judicious approach to alert

implementation since go-live

  • DDI alerts NOT enabled
slide-8
SLIDE 8

Lung Transplant patient 39 medication

  • rders

HEALTH DATA ANALYTICS - OCT 2019

slide-9
SLIDE 9

CDS alerts

HEALTH DATA ANALYTICS - OCT 2019

PRESCRIBER VIEW

13 alerts for DDIs 17 instances where Dr must complete an action – override +/- make a comment or cancel order (remove)

slide-10
SLIDE 10

Challenges with DDI alerts

  • 1. Standards
  • Currently no standards nationally or internationally
  • 2. DDI knowledgebase
  • Large variability: same DDI may be contraindicated in one, but be listed at a different severity level, or not at all

in another

  • 3. Low alert specificity = high alert burden & potential for alert fatigue
  • Alerts are not ‘context-aware’
  • Patient (lab results, co-morbidities), location, doctor, drug
  • 4. Clinical outcomes of DDI alerts
  • Evaluations in clinical practice are lacking – do DDI alerts actually prevent adverse events associated with

DDIs?3,4

HEALTH DATA ANALYTICS - OCT 2019

slide-11
SLIDE 11

HEALTH DATA ANALYTICS - OCT 2019

Drug-Drug Interaction Alerts: Help or Hindrance?

slide-12
SLIDE 12

DDI alerts – on or off?

HEALTH DATA ANALYTICS - OCT 2019

INFORMED DECISION

  • Improved functionality – enable alerts at a chosen severity level

― Limited initially – all ON or all OFF ― Moderate = 7702 pairs ― Severe = 3498 pairs ― Unknown

  • Renewed discussion

― Understand impact to prescribers ― Evidence-based data to inform decision

slide-13
SLIDE 13

To determine alert burden experienced by prescribers with existing CDS functionality, and then how this would change if DDI alerts were added to the EMM system

AIM

slide-14
SLIDE 14

Alert Conditions

Allergy & Intolerance Dose Range Local rules Therapeutic Duplication DDI

Alert Condition 1 (Live system) Reference condition

    

Alert Condition 2 (Test system)

   

(All - unknown/ moderate/severe)

  • Prescribed medication orders for all admitted patients at SVHS on a single given day were

extracted and replicated in a ‘test’ EMM system

  • The ‘Test’ system had DDI alerts enabled – unknown, moderate, severe

HEALTH DATA ANALYTICS - OCT 2019

slide-15
SLIDE 15

DDI alerts – unknown

HEALTH DATA ANALYTICS - OCT 2019

slide-16
SLIDE 16

Data entry, extraction & analysis

Live EMM system Extracted AC1 data Analyse

Input manually AC1 patient profiles

ALERT CONDITION 2

AC1 background medications

AC1 Results

Extract data using SQL queries Extract data using SQL queries

Test EMM system Extracted AC 2 data Analyse ALERT CONDITION 1

AC1 study medications

AC2 Results

EMM = electronic medication management; SQL = standard query language; AC1 = Alert Condition 1; AC2 = Alert Condition 2

HEALTH DATA ANALYTICS - OCT 2019

slide-17
SLIDE 17

RESULTS

KEY FINDINGS

slide-18
SLIDE 18

Finding 1 - Overall alert volume

Alerts generated Medication orders with at least 1 alert (%) Alerts per medication

  • rder (range)

Alert Condition 1 (No DDI alerts)

209 145 (25%) 1.4 ( 0 - 4)

Alert Condition 2 (All DDI alerts)

1063 348 (60%) 3.1 (0 - 11)

Increase

+509%* +240%* +212%*

HEALTH DATA ANALYTICS - OCT 2019

*Statistically significant increase with DDI alerts (p<0.005)

Medication orders Background: 2728 Study date: 576 Patients 254 admitted inpatients

  • 133 of these had study date medication orders
slide-19
SLIDE 19

Finding 2 - Prescriber alert burden

Alerted doctors (%) Alerts per doctor (range) Proportion of prescribed medicines that generated alerts Alert Condition 1 (No DDI alerts)

55 (71%) 3.8 (1-13) 38%

Alert Condition 2 (All DDI alerts)

71 (91%) 15 (1-85) 72%

% Increase

+121% +395% +188%

HEALTH DATA ANALYTICS - OCT 2019

576 medication orders prescribed by 78 unique doctors on the study date. Mean of 7.4 medication orders (range: 1 – 28) prescribed per doctor.

Statistically significant increase with DDI alerts (p<0.005)

slide-20
SLIDE 20

Finding 2 - Prescriber alert burden

Alerted doctors (%) Alerts per doctor (range) Alerted medication

  • rders (%)

Moderate DDIs

67 (86%) 7.8 (1- 38) 57%

% Increase +121% +205%* +145%* Severe DDIs

59 (76%) 4.7 (1-18) 32%

% Increase 107% 124% 113%

HEALTH DATA ANALYTICS - OCT 2019

*Statistically significant increase with DDI alerts (p<0.005)

slide-21
SLIDE 21

Finding 3 - DDI alert profile

Moderate 29% Severe 8% Unknown 63%

21

slide-22
SLIDE 22

WHAT DOES IT ALL MEAN?

slide-23
SLIDE 23

Finding 1 – Overall alert volume

  • With the addition of DDI alerts, almost two-thirds (60.4%) of medication orders generated an alert
  • Inpatient studies: 6.6% - 37.1%5-9

Why so high?

  • Inclusion of the ‘unknown’ DDI alerts – not common practice, low risk medicines
  • Exclusion of unknown DDI alerts  Moderate/Severe: 40%; Severe: 28%
  • Poor alert specificity – inclusion of contextual factors could reduce alert burden
  • 55% reduction in statin-drug interaction alerts if dose of statin considered the alert algorithm10
  • Commercial database – overly inclusive, prone to excessive alert generation
  • Change presentation of alerts in accordance with their severity (‘Tiering’)11
  • Relative complexity and acuity of patients at SVHS – transplant patients

HEALTH DATA ANALYTICS - OCT 2019

slide-24
SLIDE 24

28 medication

  • rders

prescribed 72% of medication

  • rders generate

an alert 20 medication

  • rders

3 alerts per

  • rder

60 alerts (vs 15 alerts)

  • For an individual prescriber, the addition of DDI alerts had a substantial impact on the number of

alerts encountered  alert fatigue

  • Unable to compare with other studies – override rates as a surrogate marker of alert fatigue12,13
  • Future work: prescriber-related outcome measures of alert burden
  • Important to consider cumulative alert burden
  • Alert volume will increase with inclusion of other CDS types in the future
  • Non-medication related alerts arising from Electronic Medical Records

Finding 2 – Prescriber alert burden

HEALTH DATA ANALYTICS - OCT 2019

slide-25
SLIDE 25

Strengths, limitations & challenges

Strengths:

  • Prescriber-focused outcome measures – burden to individual users
  • Assessed cumulative impact of enabling a specific type of CDS alert
  • Plan implementation of future CDS alert types
  • Insight into incidence of DDI alerts in an Australian hospital

Limitations:

  • One day, one EMM system, one knowledgebase, one hospital (inpatient)
  • Did not examine alert design and utility or the changes in prescribing decisions, or clinical outcomes

(e.g. reduced ADEs) Challenges:

  • Highly manual and laborious in nature – hampered scalability
  • Challenges with CDS alert data analysis +++

HEALTH DATA ANALYTICS - OCT 2019

slide-26
SLIDE 26

Looking ahead – using data to make informed decisions

  • DDI alerts need to be refined and reviewed prior to their implementation
  • Remove unknown DDI alerts, assess clinical significance,

incorporate contextual factors

  • Severe DDI alerts
  • Centralised body to develop and curate CDS content and drug

knowledgebases for an Australian context would ensure standardisation across healthcare organisations

HEALTH DATA ANALYTICS - OCT 2019

slide-27
SLIDE 27

References

1.

  • ACSQHC. Electronic medication management systems - a guide to safe implementation (3rd Edition). 3rd ed: Australian Commission on Safety and Quality in Health Care

(ACSQHC); Sydney, 2017. 2. Westbrook JI, Reckmann M, Li L, et al. Effects of two commercial electronic prescribing systems on prescribing error rates in hospital in-patients: a before and after study. PLoS Med 2012; 9(1): e1001164. 3. Wolfstadt JI, Gurwitz JH, Field TS, et al. The effect of computerized physician order entry with clinical decision support on the rates of adverse drug events: a systematic

  • review. J Gen Intern Med 2008; 23(4): 451-8.

4. Nabovati E, Vakili-Arki H, Taherzadeh Z, et al. Information Technology-Based Interventions to Improve Drug-Drug Interaction Outcomes: A Systematic Review on Features and Effects. Journal of Medical Systems 2017; 41(1): 1-17. 5. van der Sijs H, Mulder A, van Gelder T, Aarts J, Berg M, Vulto A. Drug safety alert generation and overriding in a large Dutch university medical centre. Pharmacoepidemiol Drug Saf 2009; 18(10): 941-7. 6. Kalmeijer MD, Holtzer W, van Dongen R, Guchelaar H-J. Implementation of a computerized physician medication order entry system at the Academic Medical Centre in

  • Amsterdam. Pharmacy World and Science 2003; 25(3): 88-93.

7. Zenziper Y, Kurnik D, Markovits N, et al. Implementation of a clinical decision support system for computerized drug prescription entries in a large tertiary care hospital. Isr Med Assoc J 2014; 16(5): 289-94. 8. Zenziper Straichman Y, Kurnik D, Matok I, et al. Prescriber response to computerized drug alerts for electronic prescriptions among hospitalized patients; 2017. 9. Jani YH, Barber N, Wong ICK. Characteristics of clinical decision support alert overrides in an electronic prescribing system at a tertiary care paediatric hospital. International Journal of Pharmacy Practice 2011; 19(5): 363-6. 10. Seidling HM, Storch CH, Bertsche T, et al. Successful strategy to improve the specificity of electronic statin-drug interaction alerts. Eur J Clin Pharmacol 2009; 65(11): 1149- 57. 11. Paterno MD, Maviglia SM, Gorman PN, et al. Tiering drug-drug interaction alerts by severity increases compliance rates. J Am Med Inform Assoc 2009; 16(1): 40-6. 12. Payne TH, Hines LE, Chan RC, et al. Recommendations to improve the usability of drug-drug interaction clinical decision support alerts. J Am Med Inform Assoc 2015. 13. Baysari MT, Tariq A, Day RO, Westbrook JI. Alert override as a habitual behavior - a new perspective on a persistent problem. J Am Med Inform Assoc 2017; 24(2): 409-12. 14. Phansalkar S, Desai AA, Bell D, et al. High-priority drug-drug interactions for use in electronic health records. J Am Med Inform Assoc 2012; 19(5): 735-43.

HEALTH DATA ANALYTICS - OCT 2019

slide-28
SLIDE 28

Acknowledgements

  • A/Prof Melissa Baysari
  • Prof Johanna Westbrook
  • Dr Wu Yi Zheng
  • Dr Scott Walter
  • AIHI – Macquarie University
  • St Vincent’s Hospital, Sydney – Pharmacy and IT departments
  • Dr Dennis Armstrong – DXC Technology

This research was supported by a St Vincent’s Clinic Research Grant

HEALTH DATA ANALYTICS - OCT 2019

slide-29
SLIDE 29

HEALTH DATA ANALYTICS - OCT 2019

slide-30
SLIDE 30

Finding 3 - DDI alert profile

HEALTH DATA ANALYTICS - OCT 2019

Moderate 29% Severe 8% Unknown 63%

DDI alert pair

  • No. of times DDI alert pair triggered

Severity Opioid agonists + Benzodiazepines 40 Moderate Opioid agonists + Opioid antagonists 17 Severe Opioid agonists + Various general anaesthetics 10 Moderate Opioid agonists + Pregabalin 9 Moderate Benzodiazepines + Antipsychotics 9 Severe

  • 142 unique DDI alert pairs in the data set
  • Top 5 DDI alert pairs = 27% of all DDI alerts

triggered (N= 316)

slide-31
SLIDE 31

Finding 3 - DDI alert profile

HEALTH DATA ANALYTICS - OCT 2019

  • DDI alert pair: opioid agonist + opioid antagonist = ‘severe’ interaction
  • Opioid antagonist = Targin (oxycodone + naloxone tablet)
  • Oral administration of naloxone does not produce clinically relevant systemic effects
  • Alert inappropriately generated – inclusion of contextual factors (route of administration)
  • None of the top 5 alerting pairs aligned with high priority drug interactions earmarked by the

US body co-ordinating health information technology14

slide-32
SLIDE 32

Finding 1 - Overall alert volume (AC1 vs AC2)

Alerts generated Medication orders with at least 1 alert Alerts per medication order (range)

Hospital 1 209 25% 1.4 ( 0 - 4) Hospital 2 (Moderate) 522 40% 2.3 (0 - 9) Increase

+250% +160% +164%

Hospital 2 (Severe) 277 28% 1.7 (0 - 6) Increase

+133% +113% +121%

HEALTH DATA ANALYTICS - OCT 2019

Statistically significant increase with DDI alerts (p<0.005)

slide-33
SLIDE 33

Finding 2 - Prescriber alert burden

Alerted doctors (%) Alerts per doctor (range) Alerted medication

  • rders (%)

Hospital 1

55 (71%) 3.8 (1-13) 27%

Hospital 3 (Mod/Severe)

67 (86%) 7.8 (1- 38) 57%

% Increase +121% +205% +145% Hospital 3 (Severe)

59 (76%) 4.7 (1-18) 32%

% Increase 107% 124% 113%

HEALTH DATA ANALYTICS - OCT 2019

576 medication orders prescribed by 78 unique doctors on the study date. Mean of 7.4 medication orders (range: 1 – 28) prescribed per doctor.

Statistically significant increase with DDI alerts (p<0.005)