How do I estimate the chances of survival? State of the Art Session

 

Risk assessment: The basics

(Hannah Wunsch)

Risk models usually incorporate age, co-morbidities, diagnosis, some form of vital signs (perhaps leave out physiology for simplicity)

 

Current models balance pragmatism and need for certain data; creating a good risk assessment model is hard (include a statistician!)

 

Often missing:

  • Patient preferences for care e.g DNR, choice to withdraw
  • Physician autonomy choice to withdraw
  • Support availability upon discharge e.g. family, care system

 

Small changes to things like outcome choice can have an impact

 

External validation of severity of illness models

(Hannah Wunsch)

 

External validation is important

  • describe severity illness in quantifiable way
  • knowledge of care in a hospital / region
  • ability to compare outcomes
  • assess changes in care

 

External validation will show if model performs well in other places / datasets

  • discrimination (ability of test to correctly classify those with and without disease/outcome)
  • calibration (whether or not the observed event rates match expected event rates in subgroups of the model population)

 

Don’t blame quality of care / case mix / decision-making before investigating why external validation is not working

 

Take into account any upstream issues (unrelated to ITU care itself) causing data to ‘appear different’ from expected

e.g. delay to ITU admission might mean more / better care while pt still on ward

e.g. transfer to Long Term Acute Care hospital for pt to die (hospital mortality rate looks better)

 

Consider Recalibration, and ask Why recalibration is needed… but sometimes a model truly isn’t good enough to use

 

 

Predicting Quality of Life in survivors

(Philipp Metnitz)

 

Focus on health aspects – function and feelings / wellbeing

 

>200 articles and >200 instruments published: numerous unique instruments available

– 29% articles used Baseline assessment of QOL

– 63% articles used Short Form-36; 19% used EQ-5D-3L

 

Short Form-36: patient-reported, 36 item survey covering physical / emotional / social functioning, mental health, pain, health perceptions

 

EQ-5D: health state description (mobility, self care, usual activities, pain, anxiety/depression) and evaluation using VAS

 

Physical QOL impacts on return to work, carer burden, cost to care system

 

Physicians tend to be over-optimistic in predicting QOL with approx. 30% error

IMG_0230

 

Long term cognitive impairment in ICU survivors

(Arjen Slooter)

 

Cognitive impairment after intensive care unit admission: a systematic review. (Wolters et al., Intensive Care Med 2013; 39:376-86)

  • 19 studies
  • heterogeneous population (elderly, ARDS, sepsis)
  • duration of follow up 2months – 13 yrs)
  • variation in tests applied
  • limitations – no premorbid baseline, no correction for educational level

 

Cognitive impairment frequent (11-62%)

  • impairment depends on Pre-admission function
  • duration of delirium related to cognitive impairment

 

Mechanisms in sepsis:

  • neuroinflammation (this persists in the older brain where microglia are pre-primed)
  • microcirculatory changes, hypotension, microthromboses/haemorrhages, endothelial dysfunction
  • increased NO activity
  • viral reactivation through immunosuppression (CMV / HSV)

 

Assessment currently:

  • Neuropsychological test with neuropsychologist in clinical setting (artificial situation)
  • Cognitive complaints (subjective based on emotional state, may not relate to ADL impairment ß most important aspect!)

 

Future assessment:

  • Functional testing e.g. shopping in supermarket using Virtual Reality (requires language, attention, memory, executive function)
  • VR can also be applied to training in tasks à directly relevant to patient’s ADLs

 

 

Predicting outcome one week after admission

(Dylan W. de Lange)

 

Assessing severity of illness with scoring systems: is the trend of repeated scores over time more relevant than the magnitude of a single score?

 

Designing a model for ICU pts staying > 7 days to predict 1-year mortality and QOL (EuroQOL-5D form to pts post-discharge)

 

Simple (can be used bedside) vs Detailed model: helps ‘gut feeling’ about pt for shared decision making with pt and family (but not very helpful for individual prognostication)

IMG_0263

IMG_0265