A systematic review of the magnitude and cause of geographic variation in unplanned hospital admission rates and length of stay for ambulatory care sensitive conditions

Substantial geographic variation in unplanned ACSCs admission rates and LOS is commonplace in high income countries. A broad literature exists on the topic however, to date, most research has concentrated on pneumonia, heart failure and COPD and been based in the US, UK or Canada. Primary care quality and secondary care access were often cited as drivers of admission rate variation whilst secondary care quality and the use of clinical guidelines were often listed as contributors to LOS variation. Few studies went further and empirically examined the correlates of unplanned admission rates and LOS, results from those that did suggest that increased bed availability and the absence of clinical guidelines may be an important determinant of high admission rates and LOS respectively. Several different methods were used to quantify and report variation, some studies failed to control for confounders and used naive statistical methods which limited their utility.

To our knowledge, this is the first study to systematically review the literature on geographical variation in unplanned ACSC admissions and LOS. Due to resource constraints only a sample of 300 titles and abstract decisions were checked by a second reviewer and 80 % of the full text data extraction was undertaken by a single reviewer which could reduce accuracy. However the high inter-reviewer agreement on title and abstract screening and full-text extraction suggests that our results would not have changed markedly if double screening and extraction of all studies had been undertaken. Our electronic search strategy included only Medline and Embase databases which may have missed some important studies; however our comprehensive Web of Science citation search, which included both forward and backward citations, should limit the effect of this weakness. As many of the terms used to describe geographic variation are non-specific it was difficult to identify relevant articles which may have impaired both the sensitivity and specificity of the electronic search. We have provided a guide on the magnitude of variation present in each study by categorising the authors’ conclusions as ‘significant variation’, ‘variation exists’ and ‘insignificant variation’ however these classifications are subjective.

A broad international literature exists examining variations in other facets of health care utilisation. Much of the this has built on work by Wennberg and colleagues who, since their seminal paper in 1984 [29], have developed the Dartmouth Atlas [30] in the US which has demonstrated large geographical differences in health care spending, drug use, surgical procedures and care quality over the last 30 years. Other countries have followed with their own Atlases, including England [31, 32] and Spain [33], demonstrating wide levels of variation across their healthcare systems.

Despite the overwhelming evidence that significant geographic variations exist in ACSC admission rates and LOS, relatively little is known about their cause. What little evidence that does exist is in broad agreement with our findings. Secondary care access has been widely documented [13, 34] as an important factor affecting unplanned admission rates. Evidence investigating the link between primary care quality and unplanned admission rates is mixed; two UK based studies found small and inconsistent effects [35, 36] while another found a strong association between high quality and lower cardiovascular admission rates [37]. A recent study also reported substantially decreased admission rates for ACSC conditions where GPs had financial incentives aimed at improving the quality of care [38]. Several studies have demonstrated that clinical guidelines [39] and pathways are effective modifiers of LOS.[40, 41] The effect of secondary care quality on LOS is less clear cut, while some quality improvement initiatives, such as reducing hospital acquired infections,[42] will undoubtedly lead to shorter LOS in some cases higher quality care may necessitate longer LOS [43] leading some to argue that an ‘inverted U’ association between quality and LOS is likely [44], Although issues at the interface between secondary and community care has been highlighted as an important contributor to LOS variation [6‐8], few of the studies in our review acknowledged this as an important issue or attempted to measure any association.

The substantial geographical variations displayed among the majority of studies in this review should be of concern for policymakers. The large differences, often after adjustment for a range of potential confounders, could be a marker of variable quality care and represent inefficient use of resources. Solutions are not easy, particularly as the optimal unplanned admission rate or LOS is unclear; although it is tempting to assume that lower admission rates and LOS are desirable, it may be that units with very low admission rates or LOS fail their patients if they are denying care to those who would benefit from inpatient treatment or discharging patients inappropriately early, which might result in higher readmission rates downstream [45, 46].

The key question of what can be done to reduce practice variation in potentially avoidable unplanned admissions and LOS is unlikely be answered using observational data however our results alongside the plethora of interventions tested in the wider literature [11] provide some insight. Self-management programmes have been shown to reduce unplanned admissions for patients with COPD [47] and asthma [48] while greater continuity of care with a family physician has been reported to reduce admissions rates [49]. In secondary care, early review by senior clinicians in emergency departments could be effective [50] while the association between increased local bed supply and increased admission rates [34, 23, 24] hints at supply sensitive care [13]. Closer integration between health and social care has shown promising results [51] and has been highlighted as a key aim of the UK government with the introduction of the Better Care Fund [52]. Other interventions such as telemedicine or case management have shown limited or mixed results [11]. Meanwhile, those aiming to curb unnecessary high LOS could introduce secondary care quality improvement initiatives such as clinical pathways [40, 41] or guidelines [39, 28, 27].

]The causes of variation specified by authors also serve to highlight the inherent difficulties with conducting and interpreting studies comparing process measures across several geographical units. Demographic and case-mix differences between units are common problems which require relatively complex statistical models and rich datasets to properly address. Failure to do this, as was the case with several studies included in this review, could lead to biased results. Frustratingly, several studies investigated the effect of case mix variables and, despite finding large associations, still presented unadjusted geographical comparisons [26, 53, 22, 54, 55, 28].

There were wide inconsistencies in the methods used to quantify and report variation. Most studies used graphs, maps or tables without any summary statistic quantifying the extent of variation which does not allow easy comparison between studies. Where summary measures were reported they were often statistically naïve and subject to extreme sampling variability, for example the range and coefficient of variation.

For future empirical work on variations in ACSC admissions and length of stay to be most useful for policy makers it should fulfil a number of criteria: 1) Compare a large number of regions/units to explore the factors associated with admissions and LOS; 2) Adjust for demography and case mix (prevalence/severity) and acknowledge the potential for unmeasured covariates to introduce bias; 3) Explore other, potentially modifiable, causes of variation (e.g. primary care quality, discharge planning) which might inform policy; 4) Provide a summary measure of variation (e.g. SCV) which delineates random and systematic components of variation; 5) Link to outcomes (e.g. inpatient or 90 day mortality) to make sure that the focus remains on patient outcomes rather than just the cost of the process (fewer admissions, shorter LOS).