Case 2006-MDC23-01 Proposal for rehabilitation groups in NordDRG
|Target version:||Expert Group 2007|
|MDC:||MDC23||Owner / responsible:|
|Target Grouper:||COMMON||Old forum status:||CLOSITEM - Closed item|
Last updated: 2007-04-12
Status: CLOSITEM - Closed item
Initiated - 2005
By: Nordic Working Group for Rehabilitation
Expert Network 2006-03-22 - Postpone recommended
Steering Group 2006-04-19 - Recommendation accepted
Expert Network 2007-03-28 - Change recommended
Steering Group 2007-04-11 - Recommendation accepted
Technical description of model selection
The attached set of Excel tables () describes the results of the working group analysis on Finnish rehabilitation data. The decision process is also described as a decision tree. The data comes from a number of Finish rehabilitation units participating in the Finnish FIM-project. Most of the participating units are specialized in rehabilitation. Military veteran's rehabilitation which is not based on medical needs is excluded from the data. The data was given for this development purpose by Qualisan Oy who collects it from the whole country. The data is unidentified but it is given for this purpose only and may not be used for any other purpose. The costs of care are not available and therefore length of stay is used as a proxy for it in all analyses.
The last table is the total data of 7351 cases. It can be filtered for different factors and the first line gives number of cases, mean and variation for the defined group.
Because of clinical needs the data was divided to 11 diagnostic groups for different rehabilitation situations. These groups are listed in the first table that indicates the diagnosis that belongs to each group.
The first step of the analysis was then simple linear regression explaining length of stay. The results are similar to the final results but the values do not fit in to the current structure of NordDRG. This analysis already showed that age that originally is an important explaining factor has no additional information when functional status is included in the model. Thus age needs not be taken in account in the grouper.
Data on FIM-items was dichotomize with dummy variables on 3 cut-points: greater than 2, greater than 5 and greater than 6. Thus the first dummy divides the data to two groups: FIM score for the item 1-2 (value for the dummy is 0) and 3-7 (value for the dummy is 1). The others create similar divisions but on different levels. The resulting groups are obviously partly overlapping. The data was the analyzed using linear regression on length of stay with these dummy variables as explaining factors. Even age groups were tested but they were not significant in the models. We used SAS Minimum R-squared improvement (MINR) method that test the all possible models in the data starting from the least explaining and ending at the best model for each number of explaining variables. The number of explaining variables was limited to 6 but at the end at most 3 variables were used in the final models. The models were developed for each diagnosis group (11) separately. The SAS output will be delivered to Nordic center.
Based on these models we created groups from the data. The data was first filtered for the diagnosis group at issue. Then the groups based on the best explaining dummy were formed. For example in the case of stroke this was the FIM item ‘transfer to tub’ with cut-point at greater that 2. This means that the first group are patients with score 1-2 i.e. not able to go to tub and the other group are patients with score 3-7 i.e. able to go to tub (at least) with help. For stroke the best two variable model includes in addition to ‘transfer to tub’ ‘grooming’ with cut-point at greater than 5. However, the group not able to go to tub where not possible to be divided further because all or almost all had the same value. Thus ‘grooming’ was on applied to patient helped to tub. The best three variable models included additionally ‘wheelchair’ with cut-point at greater than 2. The next step included in the regression analysis proposed a second cut-point for grooming at greater that 2. However, the length of stay of patients not able to groom (score 1-2) is shorter than that of those that just need help. This is probably true – patients with very poor prognosis are transferred to other types of care and the rehabilitation care will be shorter. The working group decided that using this kind of ‘negative’ rule would give false message and we decided as a principle to stop at this point. Finally thus 5 groups were formed for rehabilitation of stroke patients as can be seen in the table ‘stroke’.
Proposed grouping process
Rehabilitation patients are detected by the presence of procedure performed according to rehabilitation plan. This is registered by a procedure code. Such codes need to exist in each national version of procedure classification.
Expert Network 2006-03-22
The meeting expressed some doubts about the new grouping model, especially the use of Z50-group codes as indicators for rehabilitation cases.. These codes are widely used both in Sweden and Norway without linkage to “rehabilitation plan” as proposed by the rehabilitation group. The new grouping will result in significant changes in DRG assignment compared to the current situation. Such change cannot be performed without prior economical analysis. The meeting recommended that a separate "rehab"-version of the grouper is produced for testing purposes. It should be made widely available and each country should be prepared to deliver test data, preferably with costs, from at least two locations in each country by next year. The results should be evaluated.
Expert Network 2007-03-28
The proposed change will be implemented in NordDRG 2008. The indicator for rehabilitation may be Z50-group codes (Finland, Denmark, Iceland, Estonia) or a special new procedure for rehabilitation according to documented rehabilitation plan (Sweden, Norway). Because Z50 is already used both in Sweden and Norway, it may not affect DRG assignment in these countries. The countries are responsible for imforming Nordic Centre about the correct code.
The indicator of rehabilitation according to rehabilitation plan will be ZRXX00 ‘Intervention based on rehabilitation plan’ or a code in Z50 group. A new diagnosis property 23X20 ‘Rehabilitation’ (exists as inactive property already) is given to these codes. In Swedish and Norwegian versions this property is removed from diagnosis in the Z50 group. Therefore the rehabilitation part will not be available in the Norwegian versions of NordDRG.
All NASS codes from the NASS sheet will be added to ICD versions (WHO and national). They will not conflict with any existing diagnoses. They will accepted by NordDRG as secondary diagnoses. They will not be accepted as principal diagnosis and therefore they will all assigned to diagnosis category 99M00 ‘Code not acceptable as principal diagnosis’.
In the DRGlogic table new rules are created before current rule for DRG 482 (400D4002). There will be 33 new rules.
All rules will have similar structure. Icd will be ‘+’. Mdc empty, OR ‘N’, Procro1 empty, Dgcat1 empty, Agelim empty, Compl empty, Sex empty, Secproc1 empty, Disch empty, Dur empty and Rtc 0.
In addition Dgprop1 will have value 23X20 on all these rules.
The special values of each rule are given by DRG in a separate attached table (Rehab DRG rules.xls). The order of the rules in this table is central in the grouping process and must not be changed.
The rule for current DRG 462 (423D021) will be removed.
33 new DRG's are created for rehabilitation. Two old DRG's disappear.