Homogeneity of Nursing Workload
Measured by LEP
Within AP-DRGs

The author has shown in other studies that DRGs are not very homogeneous with regard to resource consumption. The question arises as to whether nursing costs as the greatest contiguous block of costs in hospital are responsible for a significant part of the heterogeneity.

AP-DRG-12 codes and nursing hours (measured by LEP of 29,893 cases from the year 2000 from the University Hospital of Zurich (USZ) were analysed.

The number of occupied DRGs is 570. For the study, only cases in DRGs populated with more then 30 cases and not in Error DRGs where selected (untrimmed: 29,893 cases in 222 DRGs; trimmed: 28,051 [-6.2 %] cases in 222 DRGs, trimpoints = Q3 + 1.5 x IQR).

LEP is a system used in Switzerland to measure nursing workload. Nursing interventions and some patient characteristics are registrated daily and weighted by standard time. – Analyses of reduction of variance (R2) and coefficients of variation (CV) of untrimmed and trimmed data with regard to length of stay (LOS) and LEP hours per case were calculated (for all cases, for all surgical cases, for all medical cases).

The main results are shown in Table 1.

R2 with regard to LOS and LEP hours per case is low (i. e. weak) for untrimmed data and moderate for trimmed data. The proportion of cases in DRGs with CV > 1.0 is noteworthy (i. e. bad) for untrimmed data, especially for medical cases, and low (i. e. good) for trimmed data. R2 is mostly lower (i. e. weaker) and proportion of cases in DRGs with CV > 1.0 is always higher (i. e. worse) when calculated with regard to nursing workload compared with the calculation with regard to LOS.

Research into the reasons for variability of nursing workload must be intensified. It has to be shown if DRG refinement can be done more (or less) accurately by using nursing criteria in addition or instead of using secondary diagnoses.

The author showed in earlier studies that DRGs of a wide variety of provenances are not particularly homogeneous in terms of either length of stay nor costs.1 The question arises as to whether nursing – as the greatest contiguous cost factor in hospitals – causes a substantial part of this heterogeneity, and as to whether nursing is able to explain it. In a first step, this study will show the extent to which nursing costs vary within AP-DRGs.

For this analysis, the University Hospital of Zurich (USZ) provided the Z/I/M with data about the 34,485 cases from the year 2000 (with an average of 2.6 diagnosis codes). [Table 2]

The analysis considered cases which received inpatient treatment and were able to be allocated the correct and adequately populated DRGs. This means that those cases were extracted which had a length of stay of 2 days or more and which were neither in Error AP-DRGs nor in AP-DRGs with fewer than 30 cases. After application of this rule, the number of cases remaining for this study amounted to 29,893.

The patient classification system used was Version 12 of the AP-DRG system on the basis of ICD-10 codes for diagnoses and ICD-9-CM/3 codes for procedures.2

Nursing workload was measured with LEP, Version 1.3 LEP means "Leistungs­erfas­sung in der Pflege" (Performance Registration in Nursing). It is a factor type nursing workload measurement system used in approximately 70 hospitals in Switzerland. LEP Version 1 encompasses 73 standardised and 7 localised items of nursing activities and some patient attributes. Each of them is weighted by a standard time. Nursing workload is measured by "LEP hours". They are calculated by adding the time values of all items marked daily for each patient.

For the purpose of trimming, a trimpoint was calculated with regard to length of stay. It was fixed at the third quartile plus 1.5 times the distance between the first and third quartiles (LOS trimpoint = Q3 + IQR × 1.5).

The reduction of Variance (R²) was used to show the extent to which the dispersion of lengths of stay and LEP hours could be explained by grouping the cases into AP-DRGs.

The coefficient of variation (CV) was used to show the extent of the dispersion of the lenghts of stay and LEP hours within AP-DRGs.

The coefficient of variation (CV) is an indicator for the nearness of the values of a sample to their mean. The higher the coefficient of variation is, the less comparable are the cases within the group under examination.

Average length of stay amounted to 7.5 days with a coefficient of variation of 1.20. With the surgical cases, average length of stay was 8.1 days (CV = 1.10), while with the medical cases, it was 7.1 days (CV = 1.28). [Table 4]

The average nursing workload measured with LEP, Version 1, was 24.4 LEP hours with a coefficient of variation of 1.67. With the surgical cases, the average nursing workload was 29.6 LEP hours (CV = 1.44), while with the medical cases, it was 21.1 LEP hours (CV = 1.87). [Table 4]

In the year 2000, a variance reduction of 22 % was reached with regard to lengths of stay and 23 % with regard to LEP hours. After trimming, the variance reduction was 42 % with regard to lengths of stay and 40 % with regard to LEP hours. [Table 5]

The variance reductions of the surgical AP-DRGs are slightly higher in comparison with the values for medical AP-DRGs, namely 28 % as opposed to 18 % with regard to lengths of stay and 25 % as opposed to 20 % with regard to LEP hours. [Table 5]

The proportion of cases in AP-DRGs with a very high variation of nursing workload (CV > 1.0) was 22 % in the surgical cases and 52 % in the medical cases. If only high variation is considered (CV > 0.5), then the proportions were 83 % in the surgical cases and 88 % in the medical cases. [Table 6]

The exclusion of outliers reduced variation, particularly in surgical cases, which accounted for no more than 39 % of the cases in AP-DRGs with CV > 0.5 and no more than 2 % in AP-DRGs with CV > 1.0 [Table 7, left]. This trimming excluded 20.2 % of nursing days and 20.8 % of the nursing workload according to LEP.

In the medical cases, 72 % of the cases still remained in AP-DRGs with CV > 0.5 and 15 % of the cases AP-DRGs with CV > 1.0 with regard to nursing workload even after trimming [Table 7, right]. Here, trimming excluded 22.0 % of nursing days and 24.1 % of the nursing workload according to LEP.

The variance reduction achieved in the year 2000 of 22 % regarding lengths of stay and 23 % with regard to LEP hours are very low values. Statistically acceptable values for variance reductions exceed 50 %. Even the variance reduction achieved with trimmed values (42 % and 40 %, respectively) are still below this value.

The exclusion of outliers with unexpectedly long lengths of stay reduces variation. In this study, they accounted for 6.2 % of the cases. The 21.2 % of nursing days and the 22.5 % of LEP hours excluded by these cases, however, indicate that these cases accounted for a very considerable share in the work and in the yield potential. Undoubtedly, it may not be assumed that the predominant part of this work must be declared inefficient. In such cases, nursing staff have an opportunity to provide evidence of whether such unexpected workloads became or may become necessary for nursing reasons. If this is the case, then it would be necessary to include nursing criteria in the use of patient classification systems and prospective payment systems.

In principle, it may be interesting to deal primarily with the more homogeneous DRGs in further research work. Here, clinical pathways could be used, for instance. However, it will still be necessary to explore the reasons which lead to outliers.

In sum, it must be said that many AP-DRGs at the University Hospital of Zurich (USZ) only allowed for a poor reflection of the nursing workload. This means that there is an unacceptably high proportion of heterogeneous AP-DRGs. This was found not only in relation to nursing workload as measured according to LEP but also with regard to lengths of stay.

With regard to the surgical cases, this problem could be countered with a suitable definition of outliers. However, this raises the question as to how the excluded fifth part of the nursing workload should be measured, weighted (and paid).

With regard to the medical cases, there are still too many AP-DRGs with high nursing workload variations even after trimming.

This means that ways must be sought of representing these parts of costs which are not explained by AP-DRGs.