Risk stratification of acutely admitted medical patients Mikkel Brabrand October 20131
Challenges No. 1 – Admissions No. 2 – Staffing Limited at best Also internationally No. 3 – Overcrowding Too many patients Universal problem October Statistics Denmark 2013
Consequences Important decisions have to made by (inexperienced) staff under difficult and stressful working conditions Diagnoses Treatment Disposition Even resuscitation October 20133
Risk stratification or perhaps better known as prognostication October 20134
Methods of risk stratification Clinical assessmentBiochemical analysesClinical scores Most of the existing systems are of suboptimal quality 1 1 Brabrand et al. Scand J Trauma Resusc Emerg Med 2010 October 20135
Predictions by nursing staff Prediction made in 1820 (63.9 %) admissions Calibration in the large 3.1 % vs. 4.7 % Experienc e Discriminatio n Goodness of fit Overall0.823p< <5 years0.728p< –9 years0.774p= –14 years0.886p=0.13 ≥15 years0.874p=0.035 October 20136
Predictions by physicians Prediction made in 734 (25.8 %) admissions Calibration in the large 2.9 % vs. 5.8 % Experienc e Discriminatio n Goodness of fit Overall0.761p< <5 years0.748p= –9 years0.955p= –14 years0.739p=0.21 ≥15 years0.846p=0.07 October 20137
Agreement by both parties 507 (17.8 %) admissions assessed Agreement (± 5 %) on 385 (75.9 %) Calibration in the large 1.0 % vs. 2.1 % Discriminatory power Goodness of fit, p = 0.91 October 20138
Biochemistry Study 2 Prognostication using blood tests October 20139
Included systems VariablePrytherch scoreFroom scoreLoekito scoreAsadollahi score Lactate dehydrogenase Bilirubin Alkaline phosphatase Bicarbonate Alanine aminotransferase Neutrophil count proportion Urea/creatinine Urea Albumin Platelets Glucose White cell count Creatinine Potassium Sodium Hemoglobin Hematocrit Age Gender Mode of admission EndpointIn-hospital mortality Imminent deathIn-hospital mortality October
Biochemical scores Prytherch score, n = 2667 (87.6 %) Froom score, n = 606 (19.9 %) Loekito score, n = 358 (11.8 %) Asadollahi score, n = 2619 (86.0 %) Calibration in the large3.5 % vs %-2.2 % vs. 0.8 %- Goodness of fitP < P = Discriminatory power Calibration in the large3.7 % vs. 3.7 %-2.8 % vs. 2.8 %- Goodness of fit – development P = 0.59P = 0.93P = 0.79 Discriminatory power – development Goodness of fit – validation P = 0.66P = 0.009P = 1.00P = 0.47 Discriminatory power – validation October
Novel clinical score Study 3 Development and validation of a novel clinical score October
Development of the models Univariable analyses 25 % cutoff Multivariable logistic regression 5 % cutoff Interaction? Deviation from linearity? Full model Simple model 1. Blood pressure 2. Heart rate 3. Respiratory rate 4. Age 5. Temperature 6. Level of consciousness 7. Oxygen saturation 8. Glucose 9. Loss of independence October
Full model CoefficientsOdds ratios Systolic blood Pressure, mmHg per Age, years per Respiratory rate, breaths/min per Loss of Independence SaO 2 /FiO 2, %/ per Intercept-2.2 Systolic blood pressure SaO 2 /FiO 2 Age Loss of independence Respiratory rate Level of consciousness Systolic blood PressureSaO 2 /FiO 2 AgeLoss of Independence Respiratory rate Level of consciousness Glucose, temperature and heart rate October
Performance of the full model Development cohort, n = 1984 (65.1 %) 1 st validation cohort n = 2261 (79.4 %) 2 nd validation cohort n = 1966 (76.8 %) Calibration in the large2.5 % vs. 2.4 %1.7 % vs. 1.8 %4.0 % vs. 3.2 % Goodness of fitP = 0.97P = 0.75P = 0.33 Discriminatory power October
PARIS score Cutoff Systolic blood Pressure≤ 115 mmHg Age≥ 80 years Respiratory rate≥ 25 breaths/min Loss of IndependenceYes Peripheral oxygen Saturation≤ 93 % or any supplemental oxygen (FiO 2 > 21%) Risk = exp(-2.2 – * sbp * age * rr * loi – * sao 2 /fio 2 )/(1+exp(-2.2 – * sbp * age * rr * loi – * sao 2 /fio 2 )) October
Performance of the PARIS score Development cohort, n = 1984 (65.1 %) 1 st validation cohort n = 2261 (79.4 %) 2 nd validation cohort n = 1966 (76.8 %) Calibration in the large --- Goodness of fitP = 0.42P = 0.74P < Discriminatory power October
External validation John Kellett has tested the PARIS score Ireland AUROC Goodness of fit p = 0.08 Uganda AUROC Goodness of fit p = 0.27 November
Where were we? 1 Brabrand et al. Scand J Trauma Resusc Emerg Med 2010 Most existing systems have been developed using inadequate methodology or have not been externally validated 1 October
Where are we now? 1 Brabrand et al. Scand J Trauma Resusc Emerg Med 2010 Most existing systems have been developed using inadequate methodology or have not been externally validated 1 Use of biochemical scores have been externally validated We have added a new clinical score (developed correctly) We have added a new player to the field (staff) So… Not that much further October
Clinical implications of our findings October
Use on individual patients Risk score (in-hospital mortality) 10.7 % PARIS score4 (≈20 % 7-day mortality) October
Individual patients Use of scoring systems on individual patients is problematic Ethics committee of the American Society of Critical Care Medicine recommends against it October
Efficacy of the scores Clinical assessment Good, but misses some Biochemical scores Good, can become better, but misses some Clinical scores Better, but misses some How can we improve this? October
Another alternative? What if we combined a biochemical score, clinical assessment and the PARIS score? Prytherch score ≥ 0.15 Risk predicted by nurse ≥ 0.15 PARIS score ≥ 3 None of approximately 1400 patients would be missed October
Price? ≈900 would be incorrectly identified as being at risk! Sensitivity % Specificity 36.4 % PPV 3.25 % NPV % October
So, we are not there, yet! November
Thank you! They study has been financially supported by Sydvestjysk Sygehus Karola Jørgensens Forskningsfond Edith og Vagn Hedegaard Jensens Fond AB Fonden Johs. M. Klein og Hustrus Mindelegat Thank you for all the assistance Staff of Sydvestjysk Sygehus and Odense University Hospital, Jesper and Torben, Birte, Anni, Annmarie and Ida And all those I have forgotten to thank! October
Please be careful out there! October