Concept: Hospital Days - How to Measure Length of Stay (LOS)

Introduction

Over time, various methods have been used to count hospital days in MCHP research projects. The three most common methods used are:


1. Use the original length of stay for each separation falling within the year. This may overcount in small areas because of very long stays found in some hospitals.
   
   
2. Truncating the original length of stay at 365 days for each separation falling within the year.
   
   
3. Count in-year days only (hospital days that fall within the fiscal year). This requires programmers to extract information from separations in at least the next year of data. See the In-Year Hospital days - How to Calculate and Fiscal Year concept for more information.
   
   
   • See In-Year Days Calculation SAS code (internal access only) for counting in-year days.

Methods for Calculating Length of Stay (LOS)

Four separate methods of computing acute care hospital inpatient Length of Stay (LOS) were compared. Annual, directly age-sex standardized, rates of days (i.e. LOS) were calculated per 1,000 population over 10 years: 1989/90 - 1998/99. See graphs below.

The 4 methods used were:


1. LOS "As Is" - this simply used the LOS variable as was recorded on each hospital discharge claim. Thus if, for example, a person was in hospital for exactly 2 years, his/her LOS variable would equal 730 days. This method would attribute all 730 days to the year the person was discharged.
   
   
2. LOS "Over Time definition" - This method was used by Roos et al (2001). This method counted the days which were only part of the fiscal year of discharge. Thus, for all hospital claims on the 98/99 dataset, for example, we checked to see if the admission date was before April 1, 1998. If so, the LOS was re-calculated as: separation date - April 1, 1998.

   So, in the above example, if the person was admitted on April 1/97 and discharged 2 years later on March 31/99 then this method would attribute 365 days in the year 98/99. The other 365 days were NOT counted at all.

3. LOS Truncated at 365 days - this simply truncated all LOS greater than 365 at 365 days. This method has been used in some of MCHP's research. As in method 2, if one was doing a count of days over multiple years, then days that are truncated for one year's count are not picked up in the count of another year.
   
   
4. LOS "In Year" - this involved assigning the days a person was in hospital to the actual fiscal year or years the bed was occupied.

   So, in the above example, 365 days would be assigned to fiscal year 97/98 and 365 days to fiscal year 98/99. This method was used by Stewart, D. et al. (2002). It is (perhaps) the most accurate count of how many hospital bed-days were "used" in a given year, however it suffers from not being able to give a correct count for the most recent year(s) because we don't have the data for people who were admitted in the most recent year(s) and have not yet been discharged.

Another example comparing all four methods

Counting days for fiscal year 97/98 AND 98/99 and a patient was admitted sometime in 97/98 but not discharged until sometime in 98/99. This person was in hospital a total of 500 days (LOS variable = 500), with 200 of the days in fiscal year 97/98 and 300 days in fiscal year 98/99.


1. LOS "As is" method: All 500 days would be attributed to the year of discharge, 98/99.
   
   
2. LOS "over time definition" method: Given the discharge date is in 98/99, this method would only count the days which occurred during 98/99 (i.e. 300); the other 200 days would not be counted at all.
   
   
3. LOS "Truncated" method : Since their discharge date is during 98/99 this method (and the "over time definition") would look at their days as "belonging to" fiscal year 98/99 only. LOS would then be truncated at 365. Thus, count 365 days would be counted in fiscal year 98/99. The other 135 days this person was in hospital would not be counted at all.
   
   
4. LOS "In Year" method: All 500 days would be counted as they occurred: 200 in fiscal yr 97/98 and 300 in fiscal yr 98/99.

All of these methods used out-patient surgery as counting 0 days, but, LOS was also computed for each method with each out-patient surgery contributing 1 day.

Note : All of these analyses involved acute care hospital separations only - Nursing Stations, Chronic, Rehab and Out-of-Prov hospitals were excluded.

1. Rates were computed for:


• Winnipeg residents as well as Non-Wpg residents (regardless of where they were hospitalized). See example graphs below.
  
  
• LOS "As Is" and LOS "In Year" - for the residents of each RHA. (See example graphs below).
  
  
• LOS "As Is" and LOS "In Year" - for the residents of each RHA District.

2. Next, we compared LOS "As Is" and LOS "In Year" by hospital location (RHA) as opposed to all of the above analyses which looked at the patient's residence (regardless of which hospital s/he was in).

3. Finally, we computed correlation coefficients (of the 10 standardized rates) between LOS "As Is" and LOS "In Year" by RHA and RHA District.

Findings

• The LOS "As Is" and the LOS "In Year" results are very similar at the residents of Winnipeg vs. Non-Wpg level.
  
  
• The LOS "truncated at 365" rates are a little lower than the above 2 but the overall pattern over the 10 years is the same.
  
  
• The LOS "Over Time definition" results are somewhat different than the other 3 methods. This method is not recommended.
  
  
• The LOS "As Is" & "In Year" rates by residents were also very similar at the RHA level, except for Churchill and perhaps NOR-MAN.
  
  
  • The correlation between these two methods of measuring LOS was also quite large for most RHA Districts (see: correlation.coeff.xls (internal access only) ).
    
  • These two measures were also similar when comparing hospitals (i.e. disregarding where the patient lived) at the RHA level with, again, the exception of Churchill and NOR-MAN.

For all graphs:


• In-patient length of stay (LOS) rates (age-sex standardized)
  
• Acute Care Hospitalizations only
  
• Outpatient surgery = 0 days

Graphs: - click on link to open graphs ...

1. Winnipeg Residents

2. NonWpg Residents

3. North Eastman RHA

4. NOR-MAN RHA

• Original LOS variable (sometimes truncated at 365 days) and allocation of in-year days are the most appropriate methods to use; they are also very close when compared across time. In-year days are more appropriate when measuring resource use, for historical analyses, and for small-area analyses.

  Note : Manitoba Health uses the original LOS variable.

• Variables to consider in days analyses are:
  
  • Is most recent year of data to be used? Then first quarter-year data should be used, if available, and footnote included in report that the most recent year of data is undercounted and why (i.e. that stays spanning two fiscal years, from the most recent year into the next, are not picked up).
    
  • Are individual facilities/small areas to be analyzed?
    
  • If average length of stay is to be analyzed, then use the original LOS variable.
  
  
• The counting method you use will depend on your research question (i.e.: whether this is costing analysis, diagnosis/procedure-specific, or population-based).
  • Tip : Where costing analyses are broken down by individual facility, it is probably best to use allocated in-year days and deal with current-year issues.

Decisions

• Use "hospital days" or "days" as concept, noting the distinction between "length of stay" and "average length of stay" - mention measuring discharges as separate concept? Decision re: how to measure days should be dated. Specify usual way of measuring days, but note that this can differ if one is trying to compare with other jurisdictions or previous analyses.
  
  
• Recommendations for counting days:
  
  
  • Using total days in research studies is typically fine especially when the next year of data is not available and institution specific or small area analysis are not being done. The overcount of days from separations that started in a prior year balances the loss of days due to admissions not yet separated in the current year.
    • NOTE : In no event should in-year days for in-year separations be used!
    
    
  • If multiple years of data are being compared then researchers should consider using in-year days.
    
  • If small area analysis/institutional analysis is being done then in-year days should be used to avoid problems with very long stays affecting days in small institutions

Programmer / Researcher Note

LOS (length of stay) is the variable that denotes the number of days an individual spent in hospital. Because of the variable name many people refer to rates and other day count measures as Total Length of Stay. To avoid confusing this measure with average length of stay, programmers and researchers should refer to these measures as total days (TD) and average length of stay (ALOS)