Setting: 69 hospitals
For this study, we used hospital data that were registered in the National Medical Registration (Landelijke Medische Registratie, LMR). All data were provided by research Institute Prismant. In the LMR, data are available of admissions in general and academic hospitals in the Netherlands. This information includes medical data such as diagnoses and surgical procedures as well as patient specific data, including age, gender and hospital stay. The LMR is not based on DBC's but diagnoses are classified by the ICD-9 and procedures by the Dutch Classification System of Procedures. There have been no major changes to these classification systems between 1991 and 2006.
Participation in the LMR is voluntary. Until 2004, the participation percentage of hospitals to the LMR was nearly 100%. Since 2005 some hospitals (2005: 2, 2006: 11) stopped their participation to the LMR because of the introduction of a second hospital registration: the registration of DBC's. This registration is obligatory and these hospitals gave priority to the DBC-registration instead of prejudicing the LMR-registration. Despite this diminishing number of participating hospitals we decided to use the 2006 data, the most recent available.
In 2006, the total number of general and academic hospitals in the Netherlands was 96; 11 of these hospitals did not participate in the LMR and 16 hospitals participated but did not register their procedures in the LMR. We excluded both of these groups in our analysis. Sixty nine hospitals (72% of the total) did contribute to this study. The excluded hospitals did not have a specific pattern in their lengths of stay. In 2004 their combined average length of stay was the same as the combined average length of stay of the 69 hospitals that were included in our study. For this reason we assumed that the data used in this study were representative of all Dutch hospitals.
A specialty was included if it had 100 or more clinical discharges. For eleven specialties, a number of hospitals were excluded because they produced too few discharges. The number of hospitals that were excluded varied from 57 hospitals for ophthalmology (a specialty that mainly works in outpatient clinics) to 1 hospital for orthopaedic surgery.
In order to compare length of stay between hospitals we applied two adjustments:
1) Adjustment for differences in the policy of admission (clinical or day-care admission)
Dutch hospitals differ in their admission policies. In principle, there is a choice between outpatient-care, day-care and clinical admission. Outpatients are treated in outpatient departments, where they consult a doctor, nurse or paramedic. Day-care is defined as care given in a specific centre for day-care to patients that only stay for several hours during the day (no overnight). Clinical patients are treated in the clinical department. They occupy a bed on a clinical ward and they intend to stay one or more overnight(s). Some hospitals tend to treat patients presenting for small procedures in day-care, while other hospitals have a larger threshold to treat in day-care. They tend to treat these patients on a clinical ward. If these patients are admitted in a clinical department, their (relatively short) length of stay contributes to the overall mean length of stay, while it does not if these patients are treated in day-care. Thus, hospitals with a larger threshold to treat patients in day-care more easily reach a short mean length of stay. In order to correct for this we excluded all hospital days of patients admitted on a clinical ward while they in principle could have been treated in day-care. In our study the hospital stay of these patients was analyzed separately. This is in accordance with the recommendation Hanning [6
] made to differentiate between same-day and overnight cases in benchmarking length of stay.
Admissions that could in principle have been treated in day-care were selected on the basis of the occurrence of the main procedure in day-care. We listed all day-care procedures that were performed at least 50 times in the Netherlands in 1997 in at least 5 hospitals. Clinical admissions with a main procedure that appeared on this list were counted as admissions that could in principle have been treated in day-care if they also complied with all of the following conditions:
• Non-acute admission;
• Admission not for delivery;
• Patient did not die in hospital;
• Maximum clinical length of stay of three days;
• Only one specialty was responsible during the stay (no transfer to another specialty);
• No transfer to another hospital.
The year 1997 was used as reference to ensure that admissions really could be treated in day-care and to avoid discussions between professionals. Therefore, there is a chance for underestimation.
2) Adjustment for case-mix
A valid comparison of lengths of stay requires case-mix adjustment. Therefore we computed for each hospital specialty a ratio of actual length of stay to expected length of stay. The expected length of stay was computed by Prismant. For each specialty the expected length of stay was based on the characteristics of its patients and the national mean length of stay that is associated with these characteristics[7
]. A ratio higher than one indicates that the length of stay is higher than if its patients had national length of stay rates. The following characteristics (variables) were taken into account:
• Age, divided in 5 classes: 0, 1–14, 15–44, 45–64, 65+ years;
• primary diagnosis. This is the main diagnosis that led to the admission); it includes about 1,000 diagnoses classified by the ICD9 in three digits;
• procedures, classified by the Dutch Classification System of Procedures. The procedures considered depend on the diagnosis of the patient. On average it includes five procedure groups.
Together these three parameters produced about 5 × 5 × 1,000 = 25,000 cells for which the mean length of stay is taken as the expected length of stay. An exception was made for patients with a length of stay of 100 hospital days and longer and for patients who died in hospital. For the latter two groups the expected length of stay was kept equal to the actual length of stay and consequently the ratio of actual length of stay to expected length of stay always was 1.
15th percentile hospital
In an Australian benchmark Hanning used the minimum length of stay as the standard (at state level) [6
]. Brownell used the hospital with the shortest overall length of stay to calculate the potential savings [5
]. For our calculation of the potential length of stay reduction, we used the 15th percentile hospital as the benchmark value. The 15th
percentile hospital of each specialty was determined by ranking the quotients of actual to expected length of stay of all hospitals with 100 or more discharges for each specialty. The hospital with the lowest ratio of actual to expected length of stay was identified as the hospital with the shortest length of stay. For each specialty the length of stay at the 15th
percentile hospital in this ranking was used as the standard for calculating the potential reduction of length of stay in all hospitals with a longer length of stay. For 2006, we calculated how many hospital days Dutch hospitals could have reduced if they had all been at least as efficient with their beds as the 15th
Experiences gained in our consultancy practice have shown that setting a realistic goal motivates medical specialists to reduce the length of stay. In the first years of our consultancy practice we used the minimum as the standard, but medical specialists had many problems with this approach. They continued emphasizing potential 'rest'-variation which was not standardized for. The use of the minimum as a standard discouraged them to work on improving the health care process. They saw it as an unattainable goal. By using the 15th percentile and not the minimum we captured potential rest variation which was not adjusted for.
Calculation of the potential reduction of length of stay in Dutch hospitals
To calculate the length of stay reduction that Dutch hospitals can achieve based on the results of the 15th percentile hospitals, we distinguished between hospital days that could be gained by substitution from clinical to day-care and hospital days that could be gained by treating clinical patients with a shorter length of stay.
An example for internal medicine:
• In the 69 hospitals of this study the total number of hospital days in clinic and day-care was 1,467,522;
• 215,587 patients were treated in day-care and 501 were treated in clinic only for 1 day;
• 3,965 patients were admitted in clinic for a 2-day (2,867 patients) or 3-day (1,098 patients) stays but could potentially have been treated in day-care;
• Treating them in day-care would save 2,867 + 1,098 + 1,098 = 5,063 hospital days, which is 0.3% of all hospital days in clinic and day-care combined;
• Without the (potential) day-care patients the total number of hospital days was 1,242,406, generated by 139,904 patients;
• The 15th percentile hospital had a ratio of actual to expected length of stay of 0.95. Using this ratio to all expected lengths of stay of every hospital, the total gain in hospital days could be 162,868, which equalled 11.1% of all hospital days in clinic and day-care combined.
As a result, for internal medicine the hospital days that could be gained by substitution from clinical to day-care was 0.3%. Hospital days that could be gained by treating clinical patients with a shorter length of stay amounted to 11.1%. The combined level was 11.4%.