Clinical Decision Support Systems in the Care of Hospitalised Patients with Diabetes
Dr. Krish Nirantharakumar
MBBS, MPH, MD, MFPH, MRCP
Senior Clinical Lecturer, University of Birmingham
Academic Public Health Consultant, Public Health England

• 15-20% hospitalised patients have a diagnosis of diabetes in UK and USA
• Poor glucose control in in-patients is associated with:
– Increased infection rates
– Increased mortality
– Increased length of stay (LOS)
• Inpatient audits suggest significant prescription errors and suboptimal glycaemic control
• Patients with diabetes concerned about the quality of care they receive
• CDSS is one of many initiatives recommended by ADA and
NHS Diabetes to improve care for patients with diabetes

 CPOE – Computerised Physician Order Entry
– Clinical Software application
– Designed for use by clinicians
– Write patient orders electronically rather than paper
 CDSS – Clinical Decision Support System
– “A computer system that uses two or more patient data to generate case specific or encounter specific advice”

• Computerised clinical decision support systems
(CDSS) are described as information systems designed to assist and improve clinical decision making
• Often referred to individual care but same principles apply when caring for a population
• These include:
– Electronic alerts
• Individual (I)
– allergy, abnormal blood results, need for a specific medication
• Population (P)
– Eligible patients not having received flu vaccination, diabetes patients not on
Statin etc
– Electronic triggers
• Naloxone for excess opioid prescription (I/P)
– Electronic guidelines (I)
– Electronic prediction models
(I/P)

1) Present findings of my doctoral research
2) Discuss opportunities to incorporate some of the work and other potential tools as part of
PICS
3) Get your opinion on an OpenClinical.net project around diabetes care

• CDSSs in the care of inpatients with diabetes in non-critical care setting: systematic review
• Epidemiological studies
– Missing discharge diagnostic codes for diabetes
– Association between hypoglycaemia and inpatient mortality & length of stay
– Association between foot disease and inpatient mortality & length of stay
– Hypoglycaemia in non-diabetic in-patients: clinical or criminal?
• A prediction model for adverse outcome in hospitalised patients with diabetes

• Includes
– Electronic prescriptions
• Insulin order template
• Modification to insulin order in line with guidelines
• Alerts to encourage “efficient” use of insulin
• Benefits
– Compliance to guidelines of prescriptions
– Higher use of basal bolus regimen
– Avoidance of sliding scale insulin
– Better glycaemic control
• (0.6-0.8mmol/l patient day weighted mean blood glucose)
– Possible reduction in length of stay
– Reduces errors in prescriptions

• POC blood glucose automatically transferred to central information system in real time
• Benefits
– As a marker of quality of care using “Glucometrics”
– Ability to describe glucose control in wards, institutions, different time periods...
– Improved blood glucose control
– Active case finding
– Surveillance

• Real-time case finding of patients in need of specialist input using health information system
– Out of range glucose results (lab, POC)
– Hypo / hyper
– Patients on insulin infusions
– Parenteral or enteral nutrition
– Patients at high risk of adverse outcome
(prolonged length of stay / mortality)

• CDSSs in the care of inpatients with diabetes in non-critical care setting: systematic review
• Epidemiological studies
– Missing discharge diagnostic codes for diabetes
– Association between hypoglycaemia and inpatient mortality & length of stay
– Association between foot disease and inpatient mortality & length of stay
– Hypoglycaemia in non-diabetic in-patients: clinical or criminal?
• A prediction model for adverse outcome in hospitalised patients with diabetes

• Under-reporting
– Scotland - Primary care data to hospital data – 41%
– England - Hospital Episode Statistics to
National diabetes audit – 33%
• Aim
– To estimate the frequency of missed discharge diagnostic codes for diabetes using inpatient electronic prescription
– To look at the feasibility of this approach in real-time correction
– To estimate the impact it would have on diabetes-related payments to the hospital Trust

• Linked PAS to PICS
• Definition of diabetes
PICS
– PAS E10-E14
– PICS - Anti-diabetic medication
• Excluding – Metformin alone in PCOS and short acting insulin prescriptions alone
• Cost calculation
– By adding a diabetes diagnostic code & recalculating tariff using HRG v4 software

• Statistics
– Estimating the frequency of missed diagnostic codes
• Capture recapture method for two sources
• Chao’s formula
– Association between missed diagnosis and admission characteristics
• Mixed effect logistic regression method

Results -
Patients admitted with diabetes identified through discharge diagnostic code and electronic prescription data for 2007-2010

Admissions with diabetes
Discharge diagnostic code present
Discharge diagnostic code or electronic prescription present
Estimated number of admissions with diabetes
22 412 25 118 27 000
Cumulative
Incidence
13.1% 14.7% 15.8%

• Change in HRG tariff code and payment was 1 in 8 only (347 out of 2,706)
• Loss per patient with change in tariff £550

Proposed algorithm to incorporate into electronic prescription and health information system to reduce missed discharge diagnostic codes

• Hospitalised patients with diabetes
– Determine the association of hypoglycaemia with:
• Mortality
• Length of stay
– Need for the study
• Limited evidence from non critical care setting
• Hospitalised non diabetic patients
– Estimate the frequency of non diabetic hypoglycaemia
– Determine if there is a plausible reason
– Explore feasibility of surveillance
– Need for the study
• True frequency unknown
• Useful in forensic cases
• Patient safety

Presence and severity of hypoglycaemia vs. length of stay and inpatient mortality
Lowest observed
>3.9 mmol/l
2.3-3.9mmol/l
≤2.2mmol
Length of Stay
1
1.51 (1.35-1.68)
2.33 (1.91-2.84)
Mortality
1
1.62 (1.16-2.27)
2.05 (1.24-3.38)

Presence of foot disease vs. inpatient mortality and length of stay
Length of Stay (RR) Mortality (OR)
Absence of foot disease code
Presence of foot disease code
Amputation
1
2.01 (1.86-2.16)
3.08 (2.60-3.65)
1
1.31 (1.04-1.65)
1.02 (0.56-1.85)

• CDSSs in the care of inpatients with diabetes in non-critical care setting: systematic review
• Epidemiological studies
– Missing discharge diagnostic codes for diabetes
– Association between hypoglycaemia and inpatient mortality & length of stay
– Association between foot disease and inpatient mortality & length of stay
– Hypoglycaemia in non-diabetic in-patients: clinical or criminal?
• A prediction model for adverse outcome in hospitalised patients with diabetes

• 602 deaths in two years in 13,794 patients with diabetes
• 13.4% (95CI 4.5% -
22.8%) more deaths
• 81 preventable deaths

• SMR was 111.2 compared to those without diabetes
• 11.2 % higher than non diabetic inpatients at
UHB (95%CI: 3.1% to
18%)
• Works out as 67 deaths
How accurate are these?
Are they useful?

• Length of stay at UHB
• Results suggest:
– Out of the 25,000 patients
10,000 have LOS less than or equal to expected median of non diabetic patients
– 85% of excess LOS contributed by 25% of the patients

• Can we identify patients with adverse outcome
(mortality and excessive length of stay) in the early phase of admission?
Source: Clement S et al. Diabetes Care. 2004; 27:553-591[40]

• Adverse outcome – composite outcome
• Composite outcome – Death or excessive length of stay
• Definition of “excessive” length of stay
– Calculate median LOS of non diabetic patient for a given clinical condition (ND-los)
– Excess LOS for every diabetes patient = LOS of diabetes patient – median of the ND-los for the same clinical condition
– Excessive LOS = Patients above the 75 th centile
• Model variables – Include demographic details, admission method, place of admission, presence of foot disease, insulin use and blood results

• Statistics
– Model development
• Multiple imputation
• Generalised estimate equations
• Logistic regression
– Internal validation (using bootstrap method)
• Discrimination – ability to differentiate (ROC curve, sensitivity, PPV etc)
• Calibration – observed probability Vs. predicted probability

ROC curve for model comparison & discrimination
0.00
0.25
Ideal Model ROC area: 0.810
Test Model ROC area: 0.784
0.50
1-Specificity
0.75
Pragmatic model ROC area: 0.802
Reference
1.00

Cut off point for the probability of having an adverse outcome
Sensitivity Specificity PPV NPV LR+ LRCorrect
0.05
0.1
0.15
0.2
0.99
0.95
0.89
0.83
0.09
0.29
0.47
0.60
0.29
0.34
0.39
0.44
0.95
0.94
0.92
0.90
1.08
1.34
1.69
2.06
0.15
0.18
0.22
0.29
0.33
0.47
0.59
0.66
0.25
0.76
0.70 0.49 0.88 2.50 0.35
0.71
0.7
0.75
0.8
0.85
0.9
0.95
1
0.3
0.35
0.4
0.45
0.5
0.55
0.6
0.65
0.21
0.17
0.13
0.08
0.04
0.01
0.00
0.68
0.60
0.54
0.47
0.40
0.35
0.30
0.25
0.98
0.99
0.99
1.00
1.00
1.00
1.00
0.77
0.83
0.88
0.91
0.93
0.95
0.96
0.97
-
0.81
0.84
0.87
0.88
0.90
0.97
0.53
0.58
0.62
0.66
0.69
0.73
0.76
0.79
0.76
0.76
0.75
0.74
0.73
0.73
0.72
0.86
0.85
0.83
0.82
0.80
0.79
0.78
0.77
-
11.35
13.79
17.34
19.38
24.02
87.53
2.98
3.60
4.34
5.19
5.98
7.11
8.45
9.99
0.81
0.84
0.88
0.92
0.96
0.99
1.00
0.42
0.48
0.53
0.58
0.64
0.69
0.73
0.77
0.77
0.76
0.75
0.74
0.74
0.73
0.72
0.75
0.77
0.78
0.79
0.79
0.78
0.78
0.78

0 .2
.4
Grouped observations
Ideal (at 45 degree)
.6
.8
loess smoother
1

1) Present findings of my doctoral research
2) Discuss opportunities to incorporate some of the work and other potential tools as part of
PICS
3) Get your opinion on an OpenClinical.net project around diabetes care

PICS and CDSS for diabetes
Establish who are diabetes patients
Monitor quality of glycaemic control
Centrally monitored
POC glucose measurements
Computerise d physician order entry
Foot care pathway
Identification of hypoglycaemia and hyperglycaemia
Electronic prescriptions
Prescription rules
& algorithm
Alerts
Link to guidelines
Electronic referral / blood tests ordering / discharge summary
Active case
finding:
1) Using prescriptions
2) POC blood glucose values
3) Prediction model

1) Present findings of my doctoral research
2) Discuss opportunities to incorporate some of the work and other potential tools as part of
PICS
3) Get your opinion on an OpenClinical.net project around diabetes care

Aides memoire for busy clinicians, making decisions about a specific patient, at a particular moment
– Checks, alerts, reminders;
– Prescribing and drug interactions;
– Intelligent order entry
Also
– Text snippets;
– Calculators;
– Search engines (Cimino’s “infobuttons”);

• First generation capabilities plus
• Evidence-based recommendations for care
– Based on practice guidelines
– Embedded in workflow/care pathway
– Explicitly manage clinical uncertainty and ambiguity
• Examples: triage, risk assessments, diagnoses, tests and investigations, treatments etc
• Specialised techniques: task modelling languages
– ASBRU, EON, GLIF, PROforma, …

)
Clinical guidelines
Research & reviews
Evidence manager

• 3G technologies will focus increasingly on integrated
care, addressing multidisciplinary, multi-morbidity and cross-sector care
• They will support quality throughout the patient journey
– Assist coordinated decision-making for clinicians
– Automate as much “grunt work” as possible
– Empower self-care by patients
• Still research: 3G will require
– much more flexible decision-making and careflow models
– ability to use knowledge resources from many providers

• Dr Jamie Coleman, UoB/ UHB
• Dr Tom Marshall, UoB
• Prof.John Fox, University of Oxford
• Dr Parth Narendran, UoB/UHB
• Dr Jonathan Webber, UHB
• Dr Mujahid Saeed, UHB
• Dr Amy Kennedy, UHB
• Prof. R.E. Ferner, UoB
• Prof. K.K. Cheng, UoB