Have hospital readmissions increased in the face of reductions in length of stay? Evidence from England

Highlights

• There is no definitive evidence about the relationship between hospital length of stay and subsequent readmission.

• Studies exploring the relationship have not conditioned on the probability of surviving the first admission.

• We analyse length of stay, in-hospital mortality and 28-day readmission for three conditions in England over 7 years.

• Stroke patients with shorter LoS were more likely to be readmitted, and the probability was unchanged as LoS fell over time.

• For hip replacement and hernia repair, length of stay was not associated with an increased probability of readmission.

Abstract

We assess the relationship between changes in hospital length of stay (LoS) and hospital quality, as measured by 28-day emergency readmission. We estimate regression models to analyse LoS and other factors associated with readmission for all those admitted for hip replacement (n = 496,334), hernia repair (n = 413,712) or following a stroke (n = 480,113) in England between 2002/3 and 2007/8. There were reductions in LoS over time while changes in crude readmission rates varied by condition. Given the high mortality rate for stroke, it is critical to account for the probability of surviving the initial admission when evaluating readmissions. Conditional upon survival, the probability of readmission was greater for stroke patients who originally had a shorter LoS and for hernia patients who had an overnight stay but there is no relationship between LoS and readmission for patients who had hip replacement. The evidence does not generally suggest that reductions in LoS were associated with an increased probability of emergency readmission.

Introduction

Concerns have been voiced that pressure for hospitals to reduce length of stay (LoS) may have adverse consequences on the quality of care experienced by patients. The “quicker and sicker” argument posits that if patients are discharged prematurely, in a less stable condition, they are at greater risk of subsequent readmission to hospital. Various studies have explored the relationship between LoS and readmission, most famously that by Kosecoff et al. who found some evidence to support the “quicker and sicker” argument following the introduction of prospective payment for Medicare patients in the United States [1]. Evidence from later studies is not definitive: some finding no relationship [2], [3], others that reductions in LoS were associated with increased readmissions [4], and another that longer LoS was associated with higher readmission [5].

To guard against adverse consequence of premature discharge, some jurisdictions penalise hospitals with higher than expected readmission rates [6], [7]. This requires taking account of the characteristics of patients that might be related to the probability of readmission. Such predictive factors include the patient's functional status, presence of co-morbidities, the type of procedure performed, whether there were post-operative complications [8], [9]; measures of socioeconomic status, such as poverty, education level, housing and marital status [10], [11]; and organisational characteristics of the local health-system [12]. But in a systematic review of risk prediction models for hospital readmissions, most were found to perform poorly [13], which could be due partly to the limited information in routine administrative datasets.

Even with better risk-adjustment, readmission rates have been criticised as a performance measure because they are correlated with another commonly used measure of hospital quality, namely in-hospital mortality [14]. If hospitals are more successful at ensuring that patients survive their initial admission, their readmission rates will likely be higher because the average health status of their survivors will be lower than if those most at risk of death had, indeed, died. In view of this, Laudicella et al. argue that readmission rates should be calculated conditional upon the likelihood that patients survive the initial admission [14]. By the same token, the relationship between LoS and readmission should also be estimated conditional upon survival. Previous studies have not done this, thereby potentially providing an inaccurate assessment of the relationship.

We employ this analytical approach, and explore the relationships between LoS, in-hospital mortality and readmissions. We focus on patients admitted to hospitals with one of three conditions, stroke (n = 480,113), hip replacement (n = 496,334) and hernia repair (n = 413,712), chosen because patients with these conditions differ markedly in terms of their LoS, and mortality and readmission rates. We evaluate these relationships for all patients admitted to English hospitals between the fiscal years 2002/3 through to 2007/8. This was a period when hospitals were under ever increasing pressure to reduce LoS, brought about by the phased introduction of the English version of prospective payment known as Payment by Results [15]. Receiving a fixed payment – the national tariff – for each type of patient treated, hospitals had stronger incentives to reduce the average cost of care, the most obvious strategy being to reduce LoS. Indeed, for all three conditions, there were pronounced reductions in LoS (or in the probability of staying overnight) over the period. We use 2002/3 as the first study period because this is the year prior to the introduction of PbR. In our exploration of the relationships between LoS, in-hospital mortality and readmissions we condition on the proportion of hospital income received from PbR, noting that other studies have found an association with LoS but not with mortality or readmission [16].

The remainder of the paper is organised as follows. In Section 2 we detail the methods we employ to explore the relationships between mortality, readmission and LoS. Section 3 provides details of the study dataset and Section 4 contains our empirical results. Section 5 discusses our results and conclusions.