Predicting, detecting, and responding to clinical deterioration on the wards: Is there room for improvement? Chris Bonafide, MD, MSCE Division of General Pediatrics [email protected] CENTER FOR PEDIATRIC CLINICAL EFFECTIVENESS CCEB
Dec 24, 2015
Predicting, detecting, and responding to clinical deterioration on the wards:
Is there room for improvement?
Chris Bonafide, MD, MSCEDivision of General Pediatrics
CENTER FOR PEDIATRIC CLINICAL EFFECTIVENESS
CCEB
Case
Case
• High-risk patient• Worsening vital signs• New oxygen requirement• Worsening labs• Concerned staff• Urgent interventions• Delayed transfer to ICU• Poor outcome
Outline
• What is clinical deterioration?
• What are rapid response systems?
• Who deteriorates?
• Do vital sign abnormalities precede deterioration?
• Once deterioration has been detected, are there barriers to calling for help?
• Summary
Outline
• What is clinical deterioration?
• What are rapid response systems?
• Who deteriorates?
• Do vital sign abnormalities precede deterioration?
• Once deterioration has been detected, are there barriers to calling for help?
• Summary
What is clinical deterioration?
Adapted from: Duncan H, Hutchison J, Parshuram CS. The Pediatric Early Warning System score: a severity of illness score to predict urgent medical need in hospitalized children. J Crit Care. Sep 2006;21(3):271-278.
A
B
C
Trajectories of Ward Hospitalization
Routine Care Needs
Increased Care NeedsVital Sign Changes
Cardiopulmonary ArrestAcute Respiratory Compromise
Death
A
B
D
C
Clinical Deterioration•Acute worsening of clinical status•On a trajectory toward arrest
Outline
• What is clinical deterioration?
• What are rapid response systems?
• Who deteriorates?
• Do vital sign abnormalities precede deterioration?
• Once deterioration has been detected, are there barriers to calling for help?
• Summary
What are rapid response systems?
• Hospital-wide systems designed to prevent cardiac arrest and death in ward patients by:1. Facilitating the identification of patients at risk2. Deploying an expert team to the bedside of patients
exhibiting signs of deterioration
• Due to strong support from safety organizations 2005-2010, most US hospitals have some form of rapid response system– CHOP– HUP
What are rapid response systems?
Rapid Response System
Afferent Arm (identification)
Efferent Arm(response)
Predictionof deterioration risk over time
Detection of active
deterioration
Medical emergency
team
Code blueteam
Standardized calling criteria
Early warning scores
Prognostication tools
Tools to supplement the clinical skills of nurses and physicians at the bedside
Mortality rate Cardiac arrest rate
AdultsNo significant reduction
Children21% reduction
Pooled
Chan PS, Jain R, Nallmothu BK, Berg RA, Sasson C. Rapid Response Teams: A Systematic Review and Meta-analysis. Arch Intern Med. Jan 11 2010;170(1):18-26.
better worse better worse
Pooled
Adults34% reduction
Children38% reduction
Rapid response systems: mixed results
Opportunities for rapid response system improvement
1. IDENTIFY a clinical profile of children at high risk of deterioration, and consider monitoring them more closely
2. DETECT deterioration more accurately using evidence-based tools
3. INTEGRATE detection into continuous physiologic monitoring systems
4. ELIMINATE barriers to calling for urgent assistance
Outline
• What is clinical deterioration?
• What are rapid response systems?
• Who deteriorates?
• Do vital sign abnormalities precede deterioration?
• Once deterioration has been detected, are there barriers to calling for help?
• Summary
Who deteriorates?
Rapid Response System
Afferent Arm (identification)
Efferent Arm(response)
Predictionof deterioration risk over time
Detection of active
deterioration
Medical emergency
team
Code blueteam
Standardized calling criteria
Early warning scores
Prognostication tools
Tools to supplement the clinical skills of nurses and physicians at the bedside
CHOP deterioration data0
51
01
52
02
5P
erc
ent
0 2 4 6 8 10 12 14 16 18Age
05
10
15
20
25
Pe
rcen
t
0 24 48 72 96 120 144 168 192 216 240hours after hospital admission
Age Hours after admission
Development of a predictive score to identify pediatric inpatients at risk of clinical deterioration
• Objective: To develop a predictive score for deterioration using non-vital sign risk factors
– Intended use: identifying high-risk children who should be intensively monitored
• Design: Case-control study• Setting: The Children’s Hospital of Philadelphia• Patients:
– Cases (n=141) were children who deteriorated while receiving care on a non-ICU inpatient unit
– Controls (n=423) were randomly selected
• Exposures: Complex chronic conditions, other patient factors, and laboratory studies in the 72h before deterioration
• Outcome: Clinical deterioration, defined as cardiopulmonary arrest, acute respiratory compromise, or urgent ICU transfer
• Analysis: Multivariable conditional logistic regression
Predictive scoreFinal multivariable conditional logistic regression model for clinical deterioration.Predictor Adjusted OR (95% CI) p-value Regression Coefficient (95% CI) Scorea
Complex Chronic Conditions
Epilepsy 4.36 (1.94-9.78) <0.001 1.47 (0.66-2.28) 2
Congenital/genetic defects 2.13 (0.93-4.89) 0.075 0.76 (-0.07-1.59) 1
Other Patient Factors
History of any transplant 3.01 (1.31-6.92) 0.010 1.10 (0.27-1.93) 2
Percutaneous or naso-enteral tube in preceding 24 hours
2.14 (1.29-3.55) 0.003 0.76 (0.25-1.27) 1
Age <1 year 1.86 (1.03-3.35) 0.038 0.62 (0.03-1.21) 1
Laboratory Studies
Blood culture sent to lab in preceding 72 hours 5.81 (3.29-10.28) <0.001 1.76 (1.19-2.33) 3
Hemoglobin <10g/dL in preceding 72 hours 3.01 (1.79-5.06) <0.001 1.10 (0.58-1.62) 2
Abbreviations: CI, confidence interval; OR, odds ratio.aScore derived by dividing regression coefficients for each covariate by the smallest coefficient (age<1 year, 0.62) and then rounding to the nearest integer. Score ranges from 0 to 12.
Results
Risk strata and estimated probabilities of deterioration.
Risk stratum Scores SSLR (95% CI) Probability of deteriorationa
Very low 0-2 0.39 (0.29-0.51) 0.06%Low 3-4 1.18 (0.85-1.64) 0.18%Intermediate 5-6 2.63 (1.74-3.96) 0.39%High 7-12 96.00 (13.24-696.17) 12.60%
Abbreviations: CI, confidence interval; SSLR, stratum-specific likelihood ratio.aCalculated using an incidence (pre-test probability) of deterioration of 0.15%.bSome individual scores above 7 include only cases.
Conclusions
• Identified a group of risk factors that may be useful to assess on admission and periodically during the hospitalization to identify patients who deserve more intensive monitoring for signs of deterioration
Next steps
• Domain validation and updating of score parameters using patients at the time of admission from the emergency department to predict deterioration in the first 12 hours
05
10
15
20
25
Pe
rcen
t
0 24 48 72 96 120 144 168 192 216 240hours after hospital admission
Hours after admission
Outline
• What is clinical deterioration?
• What are rapid response systems?
• Who deteriorates?
• Do vital sign abnormalities precede deterioration?
• Once deterioration has been detected, are there barriers to calling for help?
• Summary
Do vital sign abnormalities precede deterioration?
Rapid Response System
Afferent Arm (identification)
Efferent Arm(response)
Predictionof deterioration risk over time
Detection of active
deterioration
Medical emergency
team
Code blueteam
Standardized calling criteria
Early warning scores
Prognostication tools
Tools to supplement the clinical skills of nurses and physicians at the bedside
Pediatric Early Warning Scores
• Combine intermittent vital sign values into a manually-calculated composite score
• Monaghan’s Paediatric Early Warning Score• Haines’ Paediatric Early Warning Tool• Parshuram’s Bedside Paediatric Early Warning System Score• Edwards’ Cardiff and Vale Paediatric Early Warning System
– Abnormal parameters based on expert opinion– Not adequately validated– Variations of the scores above used widely
What is abnormal for hospitalized children?
• Age-based reference ranges for HR and RR– not evidence-based– vary widely between sources
• Better evidence exists for normal blood pressure in healthy children, but these ranges have not been evaluated in-hospital
Development of “expected” vital sign curves
• Objective: To develop expected HR, RR, SBP, and DBP curves using data from hospitalized children, to serve as the basis for:– In-hospital reference ranges– Vital sign-based early warning score development
• Design: Retrospective cohort study
• Setting: Cincinnati Children’s Hospital
• Data Source: Manually documented vital signs in EHR
• Patients: – Admissions to non-ICU inpatient units in 2008 (n=11,789)– Excluded age >=18, DNR or death during admission, LOS>1 year– Excluded vital sign observations that were physiologically implausible
• HR 0-300 = plausible
• Analysis: generalized additive models for location scale and shape (GAMLSS) using Box-Cox power exponential distribution
Vital sign data: HRn=542,766 obs
First set of curves
Vital sign data: HRn=542,766 obs
HR RR57 17356 13310 4933 11928 11546 132
79 high HR values from one patient hospitalized for 56 days
Single observations in patients who survived to discharge and
were not DNR
16 low HR values from one patient within a 4-
hour window
02
04
06
08
01
00h
r, n
um
eric
0 10 20 30 40time difference admission to observation as numeric
05
01
001
50h
r, n
um
eric
0 50 100 150time difference admission to observation as numeric
05
01
001
50h
r, n
um
eric
0 20 40 60 80 100time difference admission to observation as numeric
05
01
001
50h
r, n
um
eric
0 10 20 30 40time difference admission to observation as numeric
05
01
001
50h
r, n
um
eric
5 10 15 20 25time difference admission to observation as numeric
05
01
001
50h
r, n
um
eric
0 50 100 150time difference admission to observation as numeric
05
01
001
50h
r, n
um
eric
0 200 400 600time difference admission to observation as numeric
05
01
001
502
00h
r, n
um
eric
0 500 1000 1500 2000 2500time difference admission to observation as numeric
05
01
001
502
00h
r, n
um
eric
0 500 1000 1500time difference admission to observation as numeric
Addressing documentation error
• Used RR as a data integrity check– RR documented simultaneously– RR<HR– RR physiologically plausible (5-120)
Addressing Documentation Error
Single patient spikes still problematic
Ascertainment bias issues
• Clustering of extreme values– In a single patient experiencing an acute event
over a short time– In a single patient with abnormal baseline values
over the course of a long admission• Addressed by:
– Randomly selecting one HR from each 6-hour window of each patient’s admission
– Randomly selecting up to 10 of these values for each admission
Data for curve generation
Next steps for curve analysis
• Developing second set of curves with data integrity steps in place
• Validation using CHOP sample
• Will then use the z-scores for these curves to develop early warning score using vital sign data from case-control study
Opportunities to integrate detection tools into physiologic monitoring?
• Most inpatients are connected to physiologic monitors• Alarm parameters are set manually and adjusted as
needed• CHOP monitors generate ~20,000 alarms/day • Nurses are automatically paged with a generic message for
each of these alarms
• Can we identify and filter out false alarms?• Can physiologic data be combined to generate multi-
parameter alarms?• Can alarms be adaptive to recognize important within-
subject changes that may not reach pre-set alarm parameters?
HUP ICU Smart Alarms Project
• Evaluates HR, RR, SpO2, Skin Temp continuously• Evaluates BP measured at periodic intervals using a cuff • Compares monitored values to a model of normality generated
using neural networking methods applied to a training data set• Variance from data set used to evaluate probability that vital signs
are normal• Generates a status index ranging from 0 (no abnormalities) to 10
(severe abnormalities in all variables)
• Short-term median filtering for noise removal• Historic filtering for coping with missing parameters
http://www.obsmedical.com/products/hospital-patient-monitoring/visensia-central-station
Outline
• What is clinical deterioration?
• What are rapid response systems?
• Who deteriorates?
• Do vital sign abnormalities precede deterioration?
• Once deterioration has been detected, are there barriers to calling for help?
• Summary
Qualitative evaluation of the mechanisms by whichrapid response systems impact patient safety
• Objectives:
• To qualitatively determine how the identification and response components of rapid response systems impact nurse decision-making relevant to patient safety
• To identify barriers to recognizing and responding to clinical deterioration that exist despite rapid response system implementation
• Design: Qualitative study using semi-structured interviews
• Setting: CHOP
• Subjects: 27 nurses who care for children on non-ICU units
• Data Collection and Analysis:
• Audio recorded and transcribed interviews
• Coded using constant comparative methods
• Analyzed using a grounded theory approach
Theme: Despite implementation of an open access medical emergency team, some barriers to calling for
urgent assistance still exist.
• Some nurses doubted their own ability to recognize deterioration.
• Some nurses were hesitant to call for help for fear of being viewed as inadequate or unable to handle a difficult situation.
• While most nurses reported a collaborative working relationship with physicians, issues of hierarchy were discussed, with nurses reporting that physicians sometimes disagreed with their assessment of the need for urgent assistance. This prevented or delayed some nurses from calling the medical emergency team.
Barrier examples
• Medical nurse, 2-5 years experience:• I felt very uncomfortable with the patient… I was in there doing blood pressures and I don’t
even think I got to write them all down. I was doing them so frequently. She was very sick. I felt resistance from every member of the team. That made me hesitate to speak up. I did speak up several times, but then I stopped. I spoke up so many times saying, “This is not okay. I am extremely concerned.” Multiple times, but I never said, “No, that’s it.” I just didn’t take that last step…
• Medical nurse, 5-10 years experience:• We had a child on BiPap who we had tried everything to keep his sats up… and literally
nothing was working. At the 6:00 hour both me and the charge nurse were like, to the resident, we said, “We need you to do something. Can we just call the CAT team for a second opinion? Just something, maybe change the CPAP, just something.” We have had issues with this one particular one who insisted that, “He just needs some chest PT.” I insisted that I was doing chest PT for five straight hours now and I was doing it hard. I was doing it good. We just kept meeting resistance…
Next steps for qualitative study
• Stratify analysis by nursing characteristics
• Expansion to physicians to enable direct comparisons with nursing themes
Outline
• What is clinical deterioration?
• What are rapid response systems?
• Who deteriorates?
• Do vital sign abnormalities precede deterioration?
• Once deterioration has been detected, are there barriers to calling for help?
• Summary
Summary of opportunities for rapid response system improvement
1. IDENTIFY a clinical profile of children at high risk of deterioration, and consider monitoring them more closely
2. DETECT deterioration more accurately using evidence-based tools
3. INTEGRATE detection into continuous physiologic monitoring systems
4. ELIMINATE barriers to calling for urgent assistance
Thank you• Mentors/Collaborators
– Ron Keren– John Holmes– Vinay Nadkarni– Russell Localio– Richard Landis– Bob Berg– Kathryn Roberts– Fran Barg– Chris Feudtner– Alex Fiks– Rich Lin– Carrie Daymont– Pat Brady
• Research Assistants– Emily Huang– Kathleen McLaughlin– Shelby Drayton– Annie Chung– Duy-An Ho