Handling Patient Complexity in Casemix Classifications Associate Professor Janette Green Australian Health Services Research Institute (AHRSI) Sydney Business School, University of Wollongong, Australia ABF 2013 Sydney Acknowledgement This presentation is based on findings from a series of projects undertaken with AHSRI colleagues working in our subcentres – NCCC – AROC – PCOC – CHSD – CASiH Overview • • • • • • • • Patient complexity seen as complications and comorbidities (CCs) Are CCs important for ABF? How do we deal with them in AR-DRGs? Let’s do something about our PCCL!! Internationally? What about other classifications? Are they important in subacute care? Are they important in mental health? Patient complexity • • • • • An episode of care is assigned a principal diagnosis Additional diagnoses that affect the patient’s care are also coded (plus…) Comorbidity - pre-existing condition, not the main reason for the admission Complication - a condition acquired during a hospital stay Identified with ICD codes in the patient’s record • • • • • • Still the best way to describe today’s admitted patients? MDC based on principal diagnosis based on standards In reality, for patients who are more complex, it is harder to isolate a single diagnosis as the PDx Good care for these patients is more than just treating the PDx.... Cluster of diagnoses? Precedent with procs With more choices for care, greater concentration of complex patients admitted to hospitals How precise is the PDx? MH PDx – more than 20% do not see a MH team. Care type changes to create a new episode - CCs relevant for the first episode? But I digress... • • • • • Why is patient complexity important in an ABF world? More complex patients = more complex care required More complex care = higher cost care Complexity needs to be incorporated accurately in our casemix classifications It’s no secret that our population is ageing!! Result – • Increase in numbers of patients who are complex Increase in degree of complexity Sounds logical but what is the evidence? Using PCCL, an imperfect measure of CCs used in acute care, levels 0 to 4 Patient Clinical Complexity Level (PCCL) • • • • • • Assume we can isolate a PDx and be clear about CCs PCCL is used in acute admitted care and partly defines some DRGs Each additional diagnosis in a record is assigned a complication and comorbidity level (CCL) value from a matrix, if it is not found to be an exclusion (except neonates) Combination of CCLs produces a PCCL for each record 0 – 4 scores, would expect costs to increase Is PCCL doing its job? A surgical ADRG A medical ADRG When this methodology was developed... • A computer was a huge contraption living alone in an air conditioned room. You submitted a job, it would join a queue and you waited.... • Computer power and speed have increased Individual access to computers has improved Statistical methodology has advanced More data are available Is the PCCL methodology outdated? Certainly it is past time to review and update CCLs in the matrix may be inaccurate BUT • Who says the choice of levels is correct? Who says a matrix is the way to go? Who says the exclusions are right? Who says there are no interactions between CCs? Who says the way they're combined to create the PCCL is correct? Who says the impact is the same on everyone?... Different for paediatric patients, the elderly, other subgroups? What about incorporating procedures? What about function or severity measures? Is just updating existing CCL values in the matrix like applying a band aid to a broken leg? Internationally, some differences include... • Number of severity levels varies • • Limited eg to 2 in Nordic DRGs Unlimited eg Austria and Germany Type of CC from a limited list rather than using exclusions in some systems To determine the severity level The highest ranked CC determines the severity level in many systems ...combined with age, LOS and death during admission in France No secondary diagnosis is used in the Netherlands. If a patient is treated for an additional diagnosis they go to a new class. Other classifications - subacute • • • • • A recent review of AN-SNAP found CCs act as a cost driver A cost driver is not necessarily a good classification variable Principal diagnosis vs function AROC has been collecting data on complications and comorbidities separately, using lists of options The effect of CCs and of function on LOS are confounded More complex patients arrive with poorer function and – Have more potential to improve – Take longer to do so OR Can tolerate less therapy, so have shorter LOS Effect of Co-morbidities on ALOS and FIM change in 2006 Other (n=5196) Other (n=5196) Osteoarthritis (n=3511) Osteoarthritis (n=3511) Ischaemic heart disease (n=2288) Ischaemic heart disease (n=2288) Osteoporosis (n=1517) Osteoporosis (n=1517) Atrial fibrillation (n=1357) Atrial fibrillation (n=1357) Depression (n=773) Depression (n=773) CVA (n=758) Primary comorbidity reported Primary comorbidity reported CVA (n=758) Cardiac failure (n=639) Asthma (n=529) CAL/COPD (n=438) Visual impairment (n=424) Dementia (n=388) Cardiac failure (n=639) Asthma (n=529) CAL/COPD (n=438) Visual impairment (n=424) Dementia (n=388) Parkinson (n=296) Parkinson (n=296) Spinal cord injury /disease (n=280) Spinal cord injury /disease (n=280) Renal failure (n=236) Renal failure (n=236) Hearing impairment (n=190) Hearing impairment (n=190) Epilepsy (n=135) Epilepsy (n=135) Drug and alcohol abuse (n=104) Drug and alcohol abuse (n=104) Low er limb amputation (n=101) Low er limb amputation (n=101) Schizophrenia (n=41) Schizophrenia (n=41) Upper limb amputation (n=20) Upper limb amputation (n=20) -4.00 -2.00 0.00 2.00 Lower than the national average ALOS 4.00 6.00 8.00 10.00 Higher than the national average -3.00 -2.00 -1.00 Lower than the national average 0.00 1.00 2.00 3.00 4.00 Higher than the national average FIM change The question is... • • • Do additional diagnoses provide greater explanatory power after function has been taken into account? In AROC data set with roughly 94,000 records, CCs increased the average LOS from 17 to 23 days. Impact stronger for comps than for comorbs Number of CCs Average LOS 0 17 2 23 4 28 6 31 8 37 LOS in some stroke classes Stroke class High function Moderate function Low function, older Low function, younger 0 CCs 1 CC 3 CCs 5 CCs 14.1 15.2 17.2 18.9 22.0 23.6 25.5 25.0 34.2 33.5 40.8 40.1 46.6 47.8 52.2 55.0 Other classifications – mental health • • • New mental health classification(s) will be built PCCL would not be good as a first split, though there is some evidence that it explains some of the variability in cost of admitted episodes Splitting all inpatient mental health cases, regardless of diagnosis, into • PCCL mild and moderate – average cost $9,500 PCCL severe or catastrophic – average cost $21,500 Should be considered as a potential splitting variable In summary • A comprehensive review of PCCL is required in AR-DRGs • Work on subacute classification should consider CCs as an additional classification variable • Perhaps more than just PCCL... But be aware of confounding CCs should be tested along with other variables for inclusion in the future mental health classification Something has got to work!! References • NCCC (2013). Australian Refined Diagnostic Related Groups AR-DRG. Version 7.0: Definitions manual. Volume 3. National Casemix & Classification Centre, Australian Health Services Research Institute, University of Wollongong • AROC publications http://ahsri.uow.edu.au/aroc/index.html • Busse R, Geissler A, Quentin W, Wiley M (eds) (2011) Diagnosis-Related Groups in Europe. McGraw-Hill Thank you Janette Green Australian Health Services Research Institute University of Wollongong janette@uow.edu.au 02 4221 5734