DRG Workshop Belgrade, 18-22.November 2013. RM11 Classification adjustments/DRG exercises3 - LINDY...

DRG Workshop Belgrade, 18-22.November 2013.

RM11 Classification adjustments/DRG exercises3 -

LINDY AND RIC AND PEDJA

RIC AND LINDY


DRG classifications around the world


• for resource homogeneity and clinical meaningfulness

• To compensate for “classification failure”– The classification may not be able to measure

the selection by one hospital of particular types of cases

• Alternatively use additional flag for a price adjustment – eg use of ventilation, Dx – OR peer hospital flag.

Refinement of classes


Adjusting for classification limitations


• outliers and exceptional cases.• Many systems use outliers or exceptional case

adjustments • Well documented example is Victoria.• Critics call it “tinkering” or interfering with the

signals from the payment mechanism.• Others say that it adds precisions and fairness

to the payment system.• WHO IS RIGHT?

Extra categories in Funding model


• Substitution and different models of care

• When does the care type change?

• What is the optimum?

• What is the norm?

Outpatient and sub-acute caseload and DRGs


• IMPORTANT FOR CONSISTENT DATA

• ‘APPLES WITH APPLES’

Admission and discharge policies


• Particularly important where substitution with inpatient services can occur eg– Work up for a surgical admission– The rehabilitation phase of a joint

replacement– Or even complete episodes

• Payment neutral incentives

• Guidelines and clear definitions of payment rules

Outpatient activity management tools


• MODELLING, MODELLING, MODELLING

• Impact analysis

• Simulations

• Feedback and consultation – plan and goal alignment

Simulation of adjustments for funding precision


• The need to keep classes to manageable numbers

• Approaches to specifying outliers.

Outlier policies or classification changes


Clear Description of Costs

• Different health systems fund different activities eg – private providers usually include capital costs

through depreciation, while public providers often have a separate funding mechanism

– some systems exclude (or “unbundle”) some highly variable/high cost components of care, like intensive care or prostheses

• The contracted prices should match the appropriate costs– This might not happen if you adopt other countries

’ costs (eg no blood costs in Australian data).


Specifying Contract PricesUsually:

• Relative Value Score (Cost Weight) × Unit Price• Cost weights are derived empirically from

hospital Data• The unit price is negotiated

- ideally all hospitals would have same unit price.- the unit price can be modified to reflect:-

- differences in cost between groups of hospitals- efficiencies of scale (eg Victoria)

- transition arrangements ie “blend” the desired unit price with the hospital’s average cost (as in the Irish

Model and the private sector in Australia).


Other Hospital Products

• DRGs are only designed to describe acute admitted hospital episodes

• Different classifications are needed for other hospital services:– Outpatients– Long stay care– Health promotion activities– etc


How good are DRGs

• Typically DRGs explain about 25%-40% of the variation in the costs of treating patients.

• Hospitals don’t get a random sample of patients. Referral patterns and role delineation means that some hospitals treat sicker patients.

• Most systems have rules that provide extra payments so hospitals aren’t disadvantaged (ie share financial risk).

• With appropriate risk mitigation, funding models can explain over 80% of the variation in cost.


Rules to moderate financial risk

• Financial risk moderation for individual patients through outlier policy

• Same day policy

• Severity co-payments for specific subgroups within DRGs

• Grants

• High cost patient “adjustment funding pools”


Outlier Policy

Frequency

Length of stay (days)

Average LOS

Low Boundary

High Boundary

Paid below the standard rate

Paid the standard rate

Paid above the standard rate

High Outliers

Low Outliers

Adjustments are made to the “average” rate for patients that stay in hospital for fewer days than or more days than the pre-defined times for each DRG


General Casemix Model

Funds (red)

Days in Hospital

Low Inlier Boundary

Total $ Cost (green)

High Inlier Boundary


Same Day Policy

– Differences between the costs of DRGs in different hospitals can be due to differences in the proportions of same day cases. This can be due to:-

• Different types of cases in the DRG (COMPLEXITY)• Differences in admission/discharge policy (eg admitting rather than

treating on an outpatient basis.• Hospital clinical practice• Efficiency

– Setting separate same day payment rate within a DRG can be used to prevent hospitals with mostly overnight patients being inappropriately disadvantaged.

– Same day payment rates are not always a good idea because they can discourage hospitals from moving to same day care where appropriate – CLINICAL JUDGEMENT IS REQUIRED.


Co-payments for specific subgroups within DRGs

• Sometimes it is possible to identify subgroups within a DRG that cost more than the average. In these cases additional payments can be made.

• Co-payments are best used where groups of patients have higher than average cost in many DRGs. eg in Victoria– Mechanical ventilated patients– Native Australians

• Using too many copayments reverts back to input based funding.


Grants or ‘Block funding’

• Higher costs for some hospitals can be:– difficult to quantify (eg teaching hospital costs) or – not directly related to activity (eg running an

emergency department; the department must be kept open regardless of the level of activity)

• In such cases cash grants are often paid to hospitals in addition to activity based funding.

• Such grants are often easier to use in a public system than in a private system.


Adjustment Funding Pools• A set budget is put aside for allocating additional

funding for specific patients based upon applications from hospitals

• Example 1: New Technology in Victoria– AU$3million is set aside and hospitals apply for funding specific

technologies in small numbers of patients

• Example 2: High Cost Patient Pool in Western Australia– Approximately 20% of total hospital costs are required to treat

the most expensive 5% of patients. These patients are difficult to fund under the casemix “averaging” approach. In WA hospitals are able to apply for additional funding for individual patients, but patients’ records are independently clinically reviewed and payment approved by an industry committee.


Risk Moderation While in Transition

• When new activity based funding models are introduced not all hospitals are equally affected- some win and others lose.

• It is important to protect hospitals from extreme budgetary changes until they have time to adjust to the new funding model (ie find efficiencies).

• This is usually done by introducing “transition” grants or differential unit prices in the first few years.


Types of Casemix Models• Casemix was initially developed as a prospective payment

mechanism (ie this year’s activity determines this year’s funding). – This approach is still widely used (eg USA Medicare, Victoria

public hospitals and within the Private sector).

• Prospective payment increases the incentives to achieve technical efficiency but reduces budgetary certainty for hospitals.

• Casemix can also be used as a retrospective payment mechanism (ie last year’s activity determines this year’s funding). – This form of model is used in Ireland and New South Wales.

• Typically, in retrospective casemix models this year’s budget is set based upon last year’s budget plus growth.


Limiting the total amount of activity

• Experience suggests that health expenditure is extremely elastic and the potential to spend money on health care almost certainly exceeds any system’s capacity to pay for that care.

• Most systems attempt to limit the amount of activity funded by:– setting activity caps (ie only funding activity to a

certain level) or– excluding funding for some health care

intervention (eg cosmetic surgery) or– introducing patient contributions


Variations around activity Targets

• Hospitals cannot exactly identify how many people will be admitted

• Funding models can be designed to accommodate this uncertainty by funding activity above target activity at a marginal rate and reducing funding at a marginal rate for hospitals failing to achieve target.


Other Components of a successful casemix policy

• In the previous slides we have described the technical and process building blocks in developing a successful casemix policy.

• Non-technical issues are equally important:-– Openness and transparency– Fairness– Stakeholders involved– Formal channels of review– Willingness to listen and change


TOPICS

• USE CASES FOR DRGS AND DESIRED CRITERIA• THE GROUPING PROCESS• DESIGN OF THE DRG ALGORITHM• OVERVIEW OF NATIONAL DRG SYSTEMS• PATTERNS OF ADOPTION OF DRGs• WHERE/HOW WOULD AN I-DRG FIT IN?


MEASURING ACTIVITY LEVELS AND PAYING FOR THEM

• PbR, • ABF, • PfP, • Prospective payment,• Casemix funding, • Episode payment


GRANULARITY OF CATEGORIES– Terms – concepts

• +/- 600,000 Snomed CT terms• +/- 300,000 Snomed RT concepts

– Classification categories• +/- 15,000 Diagnoses• +/- 5,000 Procedures• 500<->1000 DRGs• 300<->400 ADRGs• [+/- 200 SRGs - +/- 100 Clinical service types]• 23 MDCs


Other -

• Care types– Rehabilitation, aged care, specialised nursing– Chronic care, Mental health.

• Service Related Groups – SRGs– Specialty utilisation measures – DRG aggeregation

• Risk adjusted capitation groupings– DCGs

• Care-staging-associated unbundled groupings – eg DBCs


DRG Design Goals

• Clinical and cost homogeneity, • Exhaustive and mutually exclusive ??????• Materiality, Transparency, • Data burden – routine clinical/admin data• Quality inputs – required precision

– clinical, – policy, and – cost


Principles of DesignGroups of healthcare activities which are:• Iso-resource

– similar resource consumption• Derived using readily available data• Clinically meaningful• Manageable number of groups• Describe actual/typical care patterns• An eye to incentives for efficiency/quality• ?? Mappable from other systems

– Benchmarking – time series


Data

• Primary data sources• Underlying classifications

– ICD/Morbidity, Procedures, Patient function

• Dependent variable– E.g. EPISODE:

• Cost, length of Stay, price, charges• Quality indicators

• Available design and test data sets


Design process

• Formal timetable of representations, Design and response– Germany, USA/Medicare (annual)

• Semi formal, biannual/annual processes– Australia, UK, Nordic

• Engagement with stakeholders– Hospitals, Clinicians, Policy, Commissioners

• Education


Statistical/classification tools• Discriminant analysis (DA)

– Uses least squares methods• Regression models (multiple and logistic)

– relationship between multiple variables• Artificial Neural Networks

– interconnected simple processors• Tree-based algorithms (CART)

– Classification and Regression Trees (CART), CHAID, AUTOGRP (Yale)

• Rules for new groups– Size, homogeneity


Clinical input and design

• Clinical Panels, representatives of medical associations– Australia, UK

• Formal representation from hospitals, medical associations– USA, Germany

• Direct clinical design input and evaluation– Practicing clinicians, full time design


Design Issues

• Purpose • Responsibilities• Design principles• Review and Revising

process• Currency, DRG unit of

activity for payment• Setting Independence• Unbundling

• Iso-Resource Groupings • Clinically meaningful• Comprehensive coverage• Readily available data• Quantitative rules • Statistical Criteria• Improving the Explanation

of Variance

USE CASE ISSUES TECHNICAL APPROACH


Grouping ProcessData cleaning and

input

Data EditsGrouping

ModellingReporting

Output

Standardise dataFixed file formatOne or multiple filesSingle patient (interactive) orMultiple patients (Batch)

Face validityConsistencyWarnings or failuresApply algorithm (s) – Table drivenUngroupables

Predictive modelsConcurrent modelsObserved v ExpectedAggregate statisticsGrouping variables

Input file & grouping variablesExpected valuesFile format(s)


Nord DRG – Respiratory grouping Logic


The full version of the manual is found at www.nordclass.uu.se


Design Structure

*APDRG, APRDRG

Major CC, Severity Scale*Major CC, Severity Scale* Major CC, Severity Scale*Major CC, Severity Scale*


Check list• CC levels, multiple levels

– Minor, Intermediate, Major + multiple• Multiple procedures

– Procedure escalators, effect of ITCs• Treatment packages

– E.g. renal dialysis, chemotherapy• Chronic care

– Stable, non-stable, catastrophic events• Generic design• Primary care

– Cross over with outpatients


Options

• Build your own DRG system• Adapting another country’s system

– (no adaptation)• International examples

– Adopt a grouper (e.g. Ireland)– Adapt a grouper (e.g. Germany)– Develop new grouper (e.g. UK, Australia)


Options (2)

• Adoption of a procedure classification– Separate decision to Casemix classification

• E.g. Germany– Joint decision

• e.g. Ireland, Portugal• International Standard Grouper

– Countries need to make decisions on grouper for domestic use (support national policies)

– Can make a separate decision for international comparisons

– Advantage in having the two related.


Clinician and other stakeholder input

• Clinical Panels, representatives of medical associations– Australia, UK

• Formal representation from hospitals, medical associations– USA, Germany

• Direct clinical design input and evaluation– Practicing clinicians, full time design


Overview of country specific variants

• Overview of country specific variants– USA Medicare’s DRGs: evolution to MS-DRGs

(Contributor: Julian Pettengill)– Australia ARDRG– Canada CMG– Germany G-DRG– England HRG– Nordic DRGs– AND OTHERS


International Evolution of DRGs

Adapted from Fetter R (1999) Casemix Classification Systems,Australian Health Review vol 22 no 2

HCFA v6-181989-2000

AP-DRGs v8-151991-98

AN-DRGs v1-31992-98

NACRICHAMPUS/DoD

1988-98

NY-DRGs v5-71988-90

APR-DRGs v8-151991-98

HCFA v5(4th Revision)

1988 (CC )exclusions

Yale RDRGs1989

HCFAVersion 1-41983- 1987

EnglandPortugalFrance

AR-DRGs v41998-2001

HRG v1-21991-1994

HRG v31997

HRG v3.52003

English Casemix Groups1989

AP-DRGs v16-231999-2006

CMS DRGs v19-232000-2006

Nord DRG1983-2006

G-DRGs v1-22003-2005

HRG v42006

Inc Non-AcuteAR-DRGs v52002IR-DRGs v1-2

1998-2003

Canada CMG1983-2006

JapanDPC

NetherlandsDBC

DRG Workshop Belgrade, 18-22.November 2013. 50

Medicare’s evolution to MS-DRGs

– In 2008 Medicare adopted Medicare-severity DRGs – From 1989 to 2007 differences in severity of illness were

captured by presence or absence of a CC– Early in the 2000s, many hospitals were beginning to take

strategic advantage of opportunities for selection:• Specialization in cardiac care and orthopedic surgery• Development of physician-owned specialty hospitals

– CMS contracted with 3M to develop MS-DRGs, which: • Expanded the number of DRGs from 500 to 750• Completely revised the CC and CC-exclusion lists• Many base DRGs are split 3 ways, with MCC, CC, no/CC


How CMS revised the CC list

• Is a given diagnosis (Dx), when present as a secondary Dx, a Major CC (MCC), a CC, or not a CC?

• Clinicians re-evaluated 13,549 Dxs to make initial MCC, CC, no CC assignments, and exclusions

• CMS measured resource impact for 3 patient groups: – The target Dx is present as a 2nd DX, and the patient has:

1. No other 2nd Dx, or all other 2nd Dxs are not CCs2. At least one other 2nd DX that is a CC, but no 2nd Dx is a MCC3. At least one other 2nd DX that is a MCC

– CMS calculated ratios of average charges for each group to average charges for all patients where no 2nd Dx is a CC.


Results

Target2nd Dx

Group 1 Group 2 Group 3

CCClassN1

Chargeratio N2

Chargeratio N3

Chargeratio

Benignhypertension 12,308 0.96 40,113 1.72 5,297 2.38

Non-CC

Obstructivebronchitis 7,003 1.42 32,276 2.19 13,355 3.04 CC

Respiratory failure 5,332 2.10 118,937 2.94 223,054 3.34 MCC


Australian Refined DRG (ARDRG)• AN-DRG v1.0

– 1992 updated annually– 1998 bianually

• AR-DRG v6, 2008• Commonwealth of Australia, Department of Health and Ageing

– Clinical Casemix Committee– National Casemix and Classification Centre (NCCC), University of Wollongong

• 23 MDCs, 665 DRGs– Surgical heirarchy, principal diagnosis

• ICD-10-AM, ACHI• Increase in groups with CC splits• http://www.health.gov.au/internet/main/publishing.nsf/content/health-casemix-ardrg1.htm


AR-DRG CC Splits

• Complication and comorbidity level (CCL)• patient clinical complexity level (PCCL)

assignment CC Level Description

0 Not a complication or comorbidity

1 Minor

2 Moderate

3 Severe

4 Catastrophic


Criteria for Partitioning groupings ARDRG v5.1

• Improved RID – partitioning should achieve a minimum 5% increase in Reduction in Deviance (RID) for LoS and total cost. Where a discrepancy occurs between the LoS and Cost analyses, total cost will take precedence as the dependent variable if the LoS distribution from the cost data can be shown to approximate the LoS distribution for the morbidity data using a chi-squared goodness-of-fit test.

• Minimum national group size – Size of DRGs created by partitioning an existing group should be the minimum of 10% of the original group and 500 estimated weighted separations (EWS). Where EWS = Estimated annual separations x 1998-99 NHCDC public sector cost weight

• Difference in resource use – New DRGs should differ in mean LoS by at least 2 days. If the longer stay group has a mean LOS of less than 4 days, then the shorter-stay group mean should not exceed 50% of the longer-stay group mean. The difference in the mean Total Cost of DRGs formed by partitioning an existing DRG should be at least 20% of the mean Total Cost of the higher cost group.

• New group homogeneity – The CV (LOS and Total Cost) for DRGs formed by partitioning and existing DRG should not exceed 1.3 x CV of the original group.

Statistical Methodology for AR-DRG Version 5.0


AR-DRG - Public Submission

• Statistical benefit• Clinical currency• Alignment with health system priorities• Implications for funding mechanisms• Maintenance of classification system stability • http://nccc.uow.edu.au/ardrg/

publicsubmission/index.html


Canada CMG• Introduced 1983• Redevelopment in late 1980s, 1990 improved• 1997, age & complexity overlays

– Age,CC not splits in CMGs– 5 comorbidity levels

• 2001, ICD-10-CA introduced• 2004 Redevelopment, 2007 CMG+

• 2007 CMG+, 2010 (CA 10th revision)– 21 MCCs, 560 CMG– 5 Factor adjustments– Age Category, Comorbidity Level, Flagged Intervention, Intervention

Event, Out-of-Hospital Intervention• CACS, Ambulatory care


CMG+ Comorbidity Levels

• Comorbidities assign patient to one of 5 Comorbidity Levels, impact on resource consumption:– Level 0 (0% to 24%)– Level 1 (25% to 49%)– Level 2 (50% to 74%)– Level 3 (75% to 124%)– Level 4 (125% or higher)


CMG+ & RIW


Germany G-DRG

• Introduced 2003, adapted from AN-DRG v4.1• Annual revision

– via the structured dialogue– 40% of all suggestions resulted in adjustments of

the weights or the classification– calculation of the relative weights of each DRG

• 2008, 1,137 DRGs, 26 Chapters– Incorporates hours on mechanical ventilation


G-DRG Functions• 16 different types of functions, global spilt criteria:

– OR-procedure not related to the principal diagnosis– Weight at admission (for patients with an age < 1 year at admission)– Specified procedures– Complex procedures– Complicating procedures– Dialysis– Polytrauma– Procedure on several locations– Intensive care therapy with a score above 552 points– Intensive care therapy with a score above 1104 points– Complex early rehabilitation therapy in geriatrics– Early rehabilitation therapy– Sequence of complex OR-procedures– Specified OR-procedures, conducted at four different time levels– Pre-transplantation hospital stay– Complicating procedures in conjunction with an allocation to the pre-MDC.


EnglandHealthcare Resource Groups (HRG)

• Payment by Results Timetable– Healthcare Resource Groups (HRG) and Service

Classification Tools (SCT)– Implementation of limited HRG Tariffs

• 2003/04 15 HRGs• 2004/05 48 HRGs (piloting of tariff)

– Tariff based system 2005/2006• Objective 60% total NHS spend• Acute Inpatients, Outpatients, A&E, Critical Care,

Mental Health


Reimbursement Issues

• Activity currency (Total Hospital Stay)• Specialised Services (Hospital Tariff Uplift)• Research and Development (Grants)• Teaching (SIFT Formula)• Critical care (per day payment)• High length of stay patients (per day payment)• Outpatients (per attendance)• Chemotherapy (Drug Costs)• High cost devices/drugs (Exception payments)


Scope of HRG4• More Settings

– Designed beyond admitted care– Allocates the same procedures to HRGs irrespective setting

• Increased Services– Chemotherapy Critical Care– Diagnostic Imaging Emergency & Urgent Care– Interventional Radiology Rehabilitation– Radiotherapy Specialist Palliative Care

• Supported by extended underlying OPCS classification (4.3)– 2,000+ new codes introduced [25% increase]– Non-surgical interventions


HRG v3.1 to v3.5 to v4

• V3.1 published 1997• V3.5 Revision October 2002 to May 2003• 19 Clinical working Groups

– Initial Meetings– Analysis Meetings– Quality Assurance of Recommendations– 572 to 610 HRGs

• HRGv4– 1398 groups (A&E, OP, Chemotherapy etc)


English Casemix Classification

• Healthcare Resource Groups

• Originally based on DRGs• Developed within the

National Health Service• Based on groups of

diagnoses and surgery/treatments

• Originally used for treatments outside of resident area

• Now used to pay hospitals for their activity via a tariff

HRG Version 1 1991

HRG Version 2 1994

HRG Version 3 1997

HRG Version 3.5 2003

HRG Version 4 2007


OTHERS

• FRANCE• AUSTRIA• JAPAN• THAILAND• COMMERCIAL – eg 3M• MALAYSIA UNU

• ??VALUE IN A COMPREHENSIVE TABULATION


International Adoption/DevelopmentCountry Diagnoses Procedures CasemixUSA ICD-9-CM ICD-9-CM, CPT4 HCFA/CMS-DRG, AP/R/APR-DRGAustralia ICD-10-AM ACHI AR-DRG, local variationsJapan ICD-10, ICF DPGSouth Korea ICD-10 KHIC-PH Korean DRGMexico ICD-9-CM ICD-9-CM HCFA DRG, version 16South Africa ICD-10 ICD-9-CM AP-DRGAustria ICD-10 Austrian LKFBelgium ICD-9-CM INAMI, /ICD-9-CM AP-DRG (1995), APR-DRG (2002)Denmark ICD-10 Nomesko Nord DRG/DAGSFinland ICD-10 NCSP Nord DRG/FinFrance ICD-10 CDAM GHM/GHJGermany ICD-10 SGVB G-DRGUK ICD-10 OPCS HRGIreland ICD-10-AM ACHI AR-DRGItaly ICD-9-CM ICD-9-CM HCFA-DRG, version 14Netherlands DBC Dutch coding lists DBC-Groups Norway ICD-10 Norwegian code Nord DRGPortugal ICD-9-CM ICD-9-CM HCFA DRGSpain ICD-9-CM ICD-9-CM HCFA DRGSweden ICD-10 Nomesko Nord DRGSwitzerland ICD-10 ICD-9-CM AP-DRGHungary ICD-10 ICPM-Hungary Hungarian grouperRomania ICD-10 ICPM AP-DRG, HCFA, IR-DRGRussia ICD-10 ICD-9-CM AP-DRG

Turkey, fyrMacedoniaCroatia,SerbiaBosnia h.Singapore

ICD-10-AMICD-10-AMICD-10-AMICD-10ICD-10-AMICD-10-AM

ACHIACHIACHIACHIACHIACHI

AR-DRG-V6AR-DRG-V6AR-DRG-V6AR-DRG-V6AR-DRG-V6AR-DRG-V6

IndonesiaMalaysiaPhilippinesMongolia

ICD-10 ?ICD-10 ?ICD-10 ?ICD-10 ?

ICD-9-cm v3ICD-9-cm v3ICD-9-cm v3ICD-9-cm v3

UNUUNUUNUUNU


TOWARDS AN INTERNATIONAL DRG?

• EURO DRG PROJECT – compatible goals?• UNU SIMPLIFIED GROUPER PROJECT –

collaboration?• APPROACHING UPTAKE OF ICD-11 AND ICHI

– NEED FOR INTERNATIOAL COMPARISONS– NEED FOR ACCESSIBLE - OPEN SOURCE MATERIAL– COST OF MAINTAINING MULTIPLE PLATFORMS -

• THE BUSINESS CASE FOR LOCAL VARIANTS


POTENTIAL GOALS

• STANDARD INTERNATIONAL CORE– At “adjacent” (general) DRG level.– Approximately 400 categories– Expandable to 800-1000 with complexity splits

• General CC tables but local levels values• Local complexity splits.• Initial scope - acute inpatients ??? but

expandable to incorporate other care types and setting independence.


Thank youDiscussion????


Statistical Evaluation .. Making it work.

• Performance of overall system– Reduction in Variance (RIV), R-Squared, &/or RAR– Log adjusted e.g. Reduction in Deviance (RID)

• Individual groups – between hospitals

– Coefficient of Variance (CV), RIV– Contribution to RIV

• Trimming– Explanation of outlier data

DRG Workshop Belgrade, 18-22.November 2013. http://en.wikipedia.org/wiki/Coefficient_of_determination

R-squared, or Coefficient of Determination


R² does not tell whether• the independent variables are a true cause of the

changes in the dependent variable;• omitted-variable bias exists;• the correct regression was used;• the most appropriate set of independent variables has

been chosen;• there is

collinearity present in the data on the explanatory variables;

• the model might be improved by using transformed versions of the existing set of independent variables.

http://en.wikipedia.org/wiki/Coefficient_of_determination


RIV with multiple factors

TSS

WSS

kn

nRAdj

112

TSS

WSS

kn

nRAdj

112

TSS

WSS

kn

nRAdj

112

TSS

WSS

kn

nRAdj

112

Glantz, SA & Slinker, BK (1990), Primer of Applied Regression and Analysis of Variance, New York, McGraw-Hill Health Professions Division.


Revision and Development

• Cycle of revision– Minor and major revisions

• Additional data sources– Inform new groups, use of other dependent variables– Inform new data collection items

• Splitting groups• Combining groups• Multivariate analyses

– Comorbidities, multiple procedures

DRG Workshop Belgrade, 18-22.November 2013. RM11 Classification adjustments/DRG exercises3 - LINDY...

Documents

Transcript of DRG Workshop Belgrade, 18-22.November 2013. RM11 Classification adjustments/DRG exercises3 - LINDY...