The results from this study confirmed the primary hypothesis that patients admitted on Friday, Saturday or Sunday stayed on average 0.31 days longer than those admitted on weekdays, after adjusting for potential confounders. However, we found that the timing of admission had an even greater effect on length of stay than initially suspected, in that those admitted on the eve of public holidays, and those admitted in the afternoons and after office hours also had a longer LOS (differences of 0.71, 1.14 and 0.65 days respectively). Of particular interest is the fact that the greatest difference in LOS was for those admitted in the afternoons – we could hypothesize that these admission episodes are missing about a day's worth of clinical work-up and management by being admitted past the optimal timing for ordering tests and procedures, which would usually be in the mornings. The above observations are hence consistent with the suspicion of clinical and front-line staff that decreased service levels available after office hours, on weekends, and during public holidays prolong LOS through delays in obtaining the necessary initial work-up for newly admitted cases. Our conclusion is similar to the study done by Iglesia and colleagues [3], who also found that weekend admission was an independent prognostic factor in determining the length of stays in patients with COPD (dichotomized as more than 3 days versus less than or equal to 3 days). On the other hand, our study results show that this phenomenon is applicable to a wider patient pool (not just COPD patients), and we have also been able to quantify the estimated difference in LOS between those admitted on weekends and weekdays, which is useful for further cost-effectiveness studies.
While the effect size reported in our study may appear small, the potential cost savings of successful programs to reduce the length of stay from admission timing should not be underestimated. For instance, if introduction of a staggered work-week could effectively reduce the LOS for the 16,994 admissions on weekends to that observed for admissions on weekdays, we could potentially save 5,207 bed-days, which would then amount to around about S$1.5 million per year in cost-savings in terms of ward-costs alone, assuming an average of S$300 per day (based on internal cost data from the TTSH Finance Department). Alternatively, a shift system allowing services to be maintained at the optimal level across a 24-hour day could have an even greater impact, as it would save about 17,600 and 13,900 bed-days in those admitted in the afternoons and after office hours respectively. This may be more difficult to implement than a staggered workweek, but could amount to greater cost-savings in excess of S$9 million per year. Although the premium for maintaining the services around the clock has to be factored into cost-benefit calculations, the above may still be underestimates of the full impact, because appropriate increments in service levels at critical timings would not just reduce LOS in the newly admitted cases, but also in other admission episodes, such as those admitted mid-week but whose stay overlaps into the weekend, for which delays may occur due to the timing-related changes in service levels.
With regards to the speciality specific analysis, it was interesting to note the differences observed between General Surgery and Orthopaedics, which are both surgical disciplines admitting acute cases, as well as between Respiratory Medicine and General Medicine, the key admitting disciplines for acute medical cases. While weekend admissions were associated with prolonged LOS for General Surgery, the effect is not borne out for Orthopaedics. One hypothesis could be that admission on a weekend delays key procedures for General Surgery patients, but not for Orthopaedics. Analysing the dates and times of operations in the two disciplines against day of admission could test such a hypothesis. Information on the type and timings of operations performed can also be extracted from routinely available administrative data and linked to admission data. This could be attempted as a follow-up study. The difference in effect between Respiratory Medicine and General Medicine, however, is more difficult to explain. One approach is to compare LOS for similar DRGs across the two disciplines, while adjusting for the potential confounders identified. Such a comparison is possible in Table 5, where two out of the top three DRGs for each specialty are common to both (DRG 177 and DRG 170). There is some suggestion that the difference between weekend and weekday admissions for DRG 177 is greater in Respiratory Medicine than in General Medicine, but more detailed sub-analysis of the data for other DRGs would be necessary before conclusions can be drawn. However, this brief sub-analysis by DRGs does generate further hypotheses about possible causes for prolonged LOS in weekend admissions – it is noted that DRG 330 (Other Gastroscopy for Major Digestive Disease W/O CC) entails procedural intervention in the form of gastroscopy, and DRG 38 (Cerebrovascular Disorders Except TIA W/O CC) often necessitates referral to paramedical services such as speech and physical therapy. Such hypotheses can be followed up through qualitative methods to identify if service levels in any specific area may be responsible for the delays, but it would appear that the delays are more specific to disease groupings rather than to disciplines.
Based on the existing analysis, we hypothesize that the service levels concerned may be broadly categorized into two types. Firstly, there is decreased medical decision-making capability, both on weekends and after office-hours. For example, ward rounds are not performed on a 24-hour basis, and there is a reduction in the ratio of senior staff to patients on weekends. The other possible bottleneck in service levels would be in the availability of critical support and paramedical services, such as investigations, allied health inputs and inter-disciplinary referrals for key procedures. A recent paper by Bell and Redelmeier [6] provides strong evidence that part of the cause does lie in the wait for procedures, with patients admitted on Fridays and Saturdays having the longest delay from admission to procedure. Bell also found that the weekend effect was procedure-specific, and this may explain the variation we observe between specialties. In all, the evidence points towards amenable causes, and a sensible approach would hence be to look for actionable factors within the specific disciplines, so as to find cost-effective ways of enhancing decision-making capability and making critical services and procedures available on a weekend.
One limitation inherent in our study was the possibility that unmeasured confounders could have affected our results. Since the source of data for this report is primarily from administrative and routine information systems, clinical data that could help us make more detailed inferences was not available. In particular, there could be concerns that the severity of the cases remains an unmeasured confounder in our study. At least two studies have found that more severe cases tend to be admitted more frequently on weekends[7, 8] However, both of these studies were based in intensive care units, whereas our study looks at all admissions, the vast majority of which are to general wards. It would be reasonable to assume that general ward admissions are less likely to be affected by fluctuations in severity by day of week, although we are unable to confirm this through the data available to us. However, we have attempted to deal with the issue of confounding by disease severity within the limitations of our dataset. Firstly, we performed a sub-analysis stratifying the data into direct admissions to ICU/HD and those initially admitted to the general ward, and found that the timing-related effects persisted at similar magnitudes in the admissions to general ward. Secondly, the sub-analysis by top DRGs (Table 5) suggests that the findings of timing-related effects are more applicable to short-stay admissions without co-morbidities and complications, and hence supports our contention that service levels rather disease severity are at work in causing this phenomenon. Lastly, we note that afternoon admissions have the longest LOS. While past studies have reported that weekend admissions may be more ill, afternoon admissions have never been noted have a greater disease severity in existing literature, and it would seem more plausible that the longer LOS in this group at least is the result of service delivery factors rather than illness severity.