- Email: [email protected]
The dynamics of the emergency medical readmission — The underlying fundamentals
EJINME-03685; No of Pages 6 European Journal of Internal Medicine xxx (2017) xxx–xxx
Contents lists available at ScienceDirect
European Journal of Internal Medicine journal homepage: www.elsevier.com/locate/ejim
Reflections in Internal Medicine
The dynamics of the emergency medical readmission — The underlying fundamentals Declan Byrne, Deirdre O'Riordan, Richard Conway, Sean Cournane, Bernard Silke ⁎ Department of Internal Medicine, St James's Hospital, Dublin 8, Ireland
a r t i c l e
i n f o
Article history: Received 5 September 2017 Received in revised form 18 September 2017 Accepted 23 September 2017 Available online xxxx Keywords: Emergency medical admission Readmissions Dynamic patterns
a b s t r a c t Background: Hospital readmissions are a perennial problem. We reviewed readmissions to one institution (2002– 2015) and investigated their dynamics. Methods: 96,474 emergency admissions (in 50,701 patients) to an Irish hospital over a 15-year period were studied, and patterns surrounding early (b 28 days) and late (any other) readmissions determined. Univariate and logistic or truncated Poisson regression methods were employed. Results: Early readmission rate averaged 9.6% (95% CI: 9.4, 9.8) with a low/high of 8.4% (95% CI: 7.8, 9.1) and 10.3% (95% CI: 9.6, 11.0) respectively with no overall time trend. Early readmissions represented 20.1% (95% CI: 19.8, 20.5) of emergency medical readmissions. Median time to first readmission was 55 weeks (95% CI: 13, 159), time to second was 35 weeks (95% CI: 9, 98); by the 7th/8th readmissions, intervals were 13 weeks (95% CI: 4, 36) and 11 weeks (95% CI: 4, 30). Readmissions were older 67.1 years (95% CI: 48.3, 79.2) vs. single admissions 53.9 years (34.3, 72.4) and stayed longer — 5.8 days (2.7, 10.6) vs. 3.9 days (1.5, 8.0). Readmissions had more Acute Illness Severity, Charlson Co-Morbidity and Chronic Disabling Disease. Between 2002 and 2015 the logistic adjusted model of 30-day in-hospital mortality reduced from 6.1% (95% CI: 5.7, 6.5) to 4.4% (95% CI: 4.1, 4.7) (RRR 30.4%). Conclusion: Early hospital readmission rate did not change over 15 years despite improvements in hospital mortality outcomes. Readmissions have a consistent pattern related to patient illness and social characteristics; the fundamentals are driven by disease progression over time. © 2017 European Federation of Internal Medicine. Published by Elsevier B.V. All rights reserved.
1. Introduction There is a rising trend of unplanned hospital medical readmissions; a particular concern are those occurring early — defined as within 1 month of hospital discharge [1]. This trend is of concern due to the questions it raises regarding the quality of care during the index hospitalisation, and also because of the overall increased burden on the acute hospital service provider [1,2]. There is the view that an early hospital readmission reflects inferior hospital care; a meta-analysis of 16 studies [3] suggested that the risk of early readmission (within 31 days) was increased by 55% with care that was judged to be of relatively low quality. Further a systematic review by Ashton et al. indicated that on average, substandard care increased the risk of early readmission by 24% [4]. However, when DesHarnais et al. [5] ranked 300 hospitals on 3 risk adjusted indices of hospital quality, mortality, readmissions and complications, there was no relationship between a hospital's ranking on any one of these indices and it's ranking on the other two. On balance the evidence suggests that readmission rates, uncorrected for confounding medical, social and hospital factors, are a poor guide to quality of care [1,6]. ⁎ Corresponding author. E-mail address: [email protected] (B. Silke).
It has been estimated that 13% of inpatients in the United States use more than half of all hospital resources through repeated admissions [7, 8]. Reported rates of unplanned emergency readmissions are 15.1% at 28 days from North East Thames [9], 28% at three months in Edinburgh [10], 38% at six months in London [11], and 19.5% at one year in Galway [12]. Overall, it has been estimated that 7% of hospital discharges result in a readmission [13], but b25% of these may be preventable [14]. The objective of this paper is to evaluate the evidence base as to whether interventions aiming to address preventing early readmissions are realistically based; we review the dynamics of readmissions with a prospectively collected database of nearly 100,000 emergency medical admissions, between 2002 and 2016, and the evidence of the data.
2. Methods 2.1. Background St James's Hospital (SJH) serves as a secondary care centre for emergency admissions for its catchment area of 270,000 adults. Emergency medical patients are admitted from the Emergency Department to an Acute Medical Admission Unit (AMAU) — the operation and outcome of which have been described elsewhere [15].
https://doi.org/10.1016/j.ejim.2017.09.028 0953-6205/© 2017 European Federation of Internal Medicine. Published by Elsevier B.V. All rights reserved.
Please cite this article as: Byrne D, et al, The dynamics of the emergency medical readmission — The underlying fundamentals, Eur J Intern Med (2017), https://doi.org/10.1016/j.ejim.2017.09.028
2
D. Byrne et al. / European Journal of Internal Medicine xxx (2017) xxx–xxx
2.2. Data collection
2.5. Statistical methods
For audit purposes we employed an anonymous patient database assembling core information about each clinical episode from elements contained on the patient administration system, the national hospital in-patient enquiry (HIPE) scheme [16], the patient electronic record, and laboratory system. Data captured includes the unique hospital number, patient demographics, principal and secondary diagnoses, procedures performed, admission and discharge dates. Additional information cross-linked and automatically uploaded to the database includes physiological and laboratory parameters. From modeling laboratory data collected at time of hospital admission we developed a predictive algorithm based on serum sodium, potassium, urea, albumin, red cell distribution width, and white blood cell count. The underlying principle is that deviation beyond the boundaries of ‘normal homeostasis’ is an estimate of risk, although the relationship is non-linear and differs for each variable. Six groups were originally defined with a 30-day mortality risk increasing in an exponential fashion [17]. We assessed the ability of known predictors — Acute Illness Severity [18,19], Charlson Co-Morbidity Index [20] and Chronic Disabling Score [21] to predict any end-points, including 30day hospital mortality, Length of Hospital stay (LOS) and any readmission (early b 28 days or any).
Descriptive statistics were calculated for background demographic data, including means/standard deviations (SD), medians/inter-quartile ranges (IQR), or percentages. Comparisons between categorical variables and mortality were made using chi-square tests. Logistic regression analysis was used to examine all significant outcome predictors (p b 0.10 from the univariate analysis) on 30-day in hospital mortality; in addition to the Acute Illness Severity Score [18, 19], Charlson Co-Morbidity Index [20] and Chronic Disabling Score [21] to predict any end-points, including 30-day hospital mortality, Length of Hospital stay (LOS) and any readmission (early b28 days or any). We then used backwards and forwards stepwise methods to determine the optimal predictors, while testing the goodness-of-fit using the Hosmer and Lemeshow's tests. The latter test is based on the expectation, that the predicted and observed frequency should match closely, and that the more closely they match, the better the fit. Adjusted Odds Ratios (OR), Relative Risk Reduction (RRR), and Number Needed to Treat (NNT) and 95% confidence intervals (CI) were calculated where appropriate. Statistical significance at P b 0.05 was assumed throughout. Stata v.13.1 (Stata Corporation, College Station, Texas, USA) was used for analysis. 3. Results
2.3. Deprivation status and calculations 3.1. Patient demographics A National Deprivation Index for Health Services Research was derived by the Small Area Health Research Unit (SAHRU) at Trinity College Dublin [22]. The census data from the Central Statistics Office (CSO) (1991, was utilized to compute a functional index of deprivation using the 3440 electoral divisions (ED's) within the Republic of Ireland; these are smallest administrative areas for which population statistics are released. A weighted combination of four indicators, relating to unemployment, social class, type of housing tenure and car ownership ranked each small area [22]. The national deprivation index scores were ranked by decile from low (least deprived) to high (most deprived) according to their ranked raw scores [23]. There are 74 EDs in the hospital catchment area with a population of 210,443 persons, as per 2006. The median population per ED was 2845 (IQR 2020, 3399). These areas were ranked nationally as Deprivation Quintile I (n = 14), Quintile III (n = 6), Quintile IV (n = 5) and Quintile V (n = 49). Using an address, data at an individual level were geo-coded and matched with the SAHRU deprivation raw score and related rank quintile. These attribute data were joined to the small area polygon geometries based upon their relative geographic positions, using the ArcGIS 10 Geographic Information System software implementation of the Pointin-Polygon algorithm, as outlined by Shimrat [24].
There were a total of 96,306 episodes recorded in 50,701 patients admitted as medical emergencies between 2002 and 2016. The proportion of males was 48.9%. The median (IQR) length of stay (LOS) was 4.7 (1.9, 9.1) days. The median (IQR) age was 58.7 (37.4, 76.0) years, with the upper 10% boundary at 86.1 yr. The major disease categories (MDC) were respiratory (25.4%), cardiovascular (16.6%), neurological (16.3%), gastrointestinal (10.2%), hepatobiliary (4.5%) and renal (4.9%). The majority of admissions were from the more socially deprived section of society (Quintiles IV & V: 89.7%). Table 1 describes the demographic characteristic of the first and subsequent readmissions. Readmissions were older 67.1 years (95% CI: 48.3, 79.2) vs. single admissions 53.9 years (34.3, 72.4) and stayed longer — readmission 5.8 days (2.7, 10.6) vs. single admission 3.9 days (1.5, 8.0). Readmissions had more Acute Illness Severity, Charlson Co-Morbidity and Chronic Disabling Disease (Table 1). Readmissions had more respiratory, but less cardiovascular and neurological MDC codings. Total readmissions (i.e. not just early b 28 days) were clearly time dependent; their cumulative proportion increased from 23.5% in 2003 to 45.9% in 2016. Overall early readmissions (b28 days) represented 9.6% (95% CI: 9.4, 9.8) of emergency medical admitted episodes. 3.2. Time pattern of readmission (Figs. 1 & 2)
2.4. Study inclusion criteria For this study, data was related to all emergency general medical patients admitted to SJH between 2002 and 2016. Each emergency medical patient was referred to the team of the ‘on-call’ Acute Medicine Consultant – ‘on-take’ for a 24 h period – most ~ 90% remained under the care of the admitting consultant for the duration of their admission. Approximately 9.9% of our patients stay N30 days with a median LOS of 54.8 days (IQR 38.8, 97.2). Consequently, the LOS data represents a highly skewed distribution. Although the clinical episode is complete for the majority by day 30, some patients remain for social reasons related to the lack of long-term care facilities. We have therefore chosen a truncated end-point (at the 30-day endpoint) for mortality analyses, to avoid these additional confounders. Readmissions obviously consider all patient episodes, irrespectively of length of hospital stay.
Between 2002 and 2016, from a total of 96,306 episodes there were 33,254 episodes with no readmission (34.5%); the readmission rate overall (excluding the first admission of a series of readmissions) was 47.6% (95% CI: 47.3, 48.0). The respective calculated readmission rates at 1, 3, 5 and 10 yr were 32.4%, 40.6%, 48.9% and 53.3% respectively. Over the 15 yr period, early readmissions (b 28 days) represented 20.1% (95% CI: 19.8, 20.5) of all readmissions. Following an emergency medical admission, the risk of a readmission is high initially but rapidly falls albeit with a long tail (Fig. 1). The best fit of the time to readmission was the beta density distribution function. The alpha 1 and 2 parameters were 0.70 and 2.17 respectively. The calculated time to readmission for the 5th, 10th, 25th, 50th, 75th, and 90th centiles were 6, 14, 48, 145, 288 and 441 weeks. The time from the first to each subsequent readmission is described (Fig. 2). The median time was calculated in weeks between the first and
Please cite this article as: Byrne D, et al, The dynamics of the emergency medical readmission — The underlying fundamentals, Eur J Intern Med (2017), https://doi.org/10.1016/j.ejim.2017.09.028
D. Byrne et al. / European Journal of Internal Medicine xxx (2017) xxx–xxx
3
Table 1 Characteristics of emergency medical admissions (2002–2016). Factor N Gender Outcome Age, median (IQR) Length of stay (days) Acute illness severity
Charlson Index
Disabling disease
Sepsis status
MDC respiratory MDC cardiovascular MDC neurology
Level Male Female Alive Died
1 2 3 4 5 6 0 1 2 0 1 2 3 4 0 1 2 0 1 0 1 0 1
No readmission
1+ readmission
30,236 14,725 (48.7%) 15,511 (51.3%) 28,328 (93.7%) 1908 (6.3%) 53.9 (34.3, 72.4) 3.9 (1.5, 8.0) 1705 (6.7%) 3395 (13.2%) 4475 (17.5%) 4738 (18.5%) 4155 (16.2%) 7168 (28.0%) 18,309 (60.6%) 6205 (20.5%) 5722 (18.9%) 5272 (17.4%) 9835 (32.5%) 8171 (27.0%) 4680 (15.5%) 2278 (7.5%) 23,129 (76.5%) 6059 (20.0%) 1048 (3.5%) 24,570 (81.3%) 5666 (18.7%) 25,116 (83.1%) 5120 (16.9%) 24,043 (79.5%) 6193 (20.5%)
56,347 27,608 (49.0%) 28,739 (51.0%) 54,287 (96.3%) 2060 (3.7%) 67.1 (48.3, 79.2) 5.8 (2.7, 10.6) 845 (1.5%) 2595 (4.6%) 5697 (10.1%) 9025 (16.0%) 12,070(21.4%) 26,115 (46.3%) 21,441 (38.2%) 17,677 (31.5%) 17,052 (30.4%) 4223 (7.5%) 11,956 (21.2%) 17,234 (30.6%) 13,627 (24.2%) 9307 (16.5%) 42,865 (76.1%) 11,517 (20.4%) 1965 (3.5%) 40,033 (71.0%) 16,314 (29.0%) 47,108 (83.6%) 9239 (16.4%) 48,447 (86.0%) 7900 (14.0%)
P-value 0.41 b0.001 b0.001 b0.001 b0.001
b0.001
b0.001 Fig. 2. The data illustrate the cumulative time (weeks) between consecutive readmissions. The interval between each readmission progressively declined and approximated to an exponential model. 0.36
3.3. Comparison of the mortality and readmissions rates over time (Fig. 3) b0.001 0.043 b0.001
each subsequent admission. A mathematical model, a one-phase exponential association, satisfactorily fits the cumulative time intervals in weeks of the first eight readmissions, with a correlation coefficient of 0.99. As time advances the duration between consecutive admissions progressively shortens. The median time to the first readmission was the longest at 55 weeks (95% CI: 13, 159), the time to the second was less at 35 weeks (95% CI: 9, 98) but by the 7th and 8th readmissions, the corresponding intervals were 13 weeks (95% CI: 4, 36) and 11 weeks (95% CI: 4, 30).
The St James' Hospital medical emergency database was set up in 2002, primarily to track the impact of a new initiative in acute medicine, namely the acute medical admission unit (AMAU). While there may be factors between 2002 and 2016 that may have improved outcome over those years, it is of interest to look at mortality and readmission rates over these years. The change in 30-day in-hospital mortality over the study period is shown in Fig. 3. Calculated by episode (ignoring readmissions) the logistic adjusted model of 30-day in-hospital mortality showed a decline from 6.1% (95% CI: 5.7, 6.5) in 2002 to 4.4% (95% CI: 4.1, 4.7); the relative risk reduction was 30.4% with a NNT of 60.1. Calculating on unique patients (one episode only counted — last if N one admission), the logistic adjusted model of 30-day in-hospital mortality showed a decline from 17.3% (95% CI: 16.4, 18.2) in 2002 to 8.0% (95% CI: 7.5, 8.5); the relative risk reduction was 61% with a NNT of 13.5. These remarkable improvements in outcome have to be compared with the readmissions over time. On average the early readmission rate (b28 days) averaged 9.6% (95% CI: 9.4, 9.8) with a low and high of 8.4 (95% CI: 7.8, 9.1) and 10.3 (95% CI: 9.6, 11.0) respectively with no overall time trend.
3.4. Importance of deprivation status and illness severity on the hospital readmission rate (Fig. 4)
Fig. 1. Time to readmission following an index emergency medical admission. The model best fit was described by the Beta density distribution with parameters alpha1 and alpha2 of 0.70 and 2.17 respectively. Time (weeks) to readmission at the 75, 50, 25 and 10th centiles was 50, 145, 288 and 432 weeks respectively.
The data clearly showed that readmissions to our acute hospital were strongly related to the Deprivation status of the patients. Over the 15 year period, we calculated the total number of readmissions from each small area (i.e. Electoral Division); relating these to each ED population allowed a readmission rate to be calculated (readmission rate / 1000 population) which in turn allowed us to calculate the incidence rate ratio (IRR). Readmission was much more likely from a high deprivation area — IRR 1.69 (95% CI: 1.45, 1.98). An eight fold (×8) variation in calculated readmission rate was identified between the least deprived group (Q1) — 2.5/1000 (95% CI: 1.24, 3.76), and the most deprived group (Q5) — 20.6/1000 (95% CI: 15.6, 25.6. There were other predictive factors assessed in the multiple variable logistic model. These included the Acute Illness Severity (18, 19); those with higher illness scores were more likely have a readmission — IRR 1.37 (95% CI: 1.27, 1.48). Age might be conjectured as a likely association with an increased readmission rate; however adjusted for illness severity and complexity this was not the case IRR 0.98 (95% CI: 0.98,
Please cite this article as: Byrne D, et al, The dynamics of the emergency medical readmission — The underlying fundamentals, Eur J Intern Med (2017), https://doi.org/10.1016/j.ejim.2017.09.028
4
D. Byrne et al. / European Journal of Internal Medicine xxx (2017) xxx–xxx
Fig. 3. Between 2002 and 2016, the logistic adjusted model of 30-day in-hospital mortality showed a per episode decline from 6.1% (95% CI: 5.7, 6.5) to 4.4% (95% CI: 4.1, 4.7) — a relative risk reduction of 30.4%. The logistic adjusted model for the early readmission rate (b28 days) averaged 9.6% (95% CI: 9.4, 9.8) of episodes with a low and high of 8.4 and 10.9 respectively with no overall time trend.
0.99). Complexity could be inferred from either the Chronic Disabling Disease Score [21] or the Charlson Index [20]. A higher Chronic Disabling Score [21] was strongly associated with more readmissions — IRR 1.20 (95% CI: 1.14, 1.26); higher Charlson scores predicted a lower readmission likelihood — IRR 0.87 (95% CI: 0.81, 0.93). The Charlson Index is weighted with cancer and cardiovascular bias — which are likely to have a worse prognosis; similarly the presence of sepsis [25], for possibly similar reasons, was less likely to be associated with further readmissions — IRR 0.81 (95% CI: 0.75, 0.87). The AUROC in the model was 0.61 (95%CI: 0.60, 0.62).
Fig. 4. The hospital readmission incidence (rate/1000 population) was strongly influenced by the deprivation status. The Irish National Index ranked the catchment small areas from low Q1 to high Q5 deprivation. The predicted probabilities were derived from the multiple variable truncated Poisson model; the effect with confidence interval is plotted based on the latter prediction (No Q2 Electoral Divisions exist within the hospital catchment area).
4. Discussion Emergency admissions and readmissions are rising in many countries [13]; healthcare organisations such as the Department of Health in England and Medicare and Medicaid in the USA have produced guidance on restricting payments for readmission within 30 days of discharge from a previous (index) admission [26,27]. The crucial question that follows is, of course, as to what is the extent to which readmissions are preventable? [28]. Our data, collected in one institution over 15 years, demonstrates that following an emergency medical admission, the risk of a readmission is high initially but rapidly declined with a long tail. On average the early readmission rate (b28 days) averaged 9.6% (95% CI: 9.4, 9.8) with a low and high of 8.4% (95% CI: 7.8, 9.1) and 10.3% (95% CI: 9.6, 11.0) respectively with no overall time trend. Blunt et al. [13] estimated that early readmissions over a 6 year period were equivalent to 7.0% of hospital discharges. However, early readmissions are part of a distribution continuum; the question not being addressed is whether early readmissions can be disassociated from late readmissions. Or more specifically whether it is realistic to intervene and attempt to selectively alter part of a distribution? Geoffrey Rose focused on this issue in a classic paper [29]; he argued in favour of the population approach (targeting the entire distribution) rather than some specific component. Or as he aptly put it, the distribution is the way it is for complex reasons that one may not comprehend and it would resist well-meaning enthusiastic interventions designed to modify one aspect of the curve. Modeling the curve, suggested that the best fit was with a beta distribution model. The beta distribution is a family of continuous probability distributions defined on the interval [0, 1] parameterized by two positive shape parameters, typically denoted by α and β. The latter control the shape of the distribution. The Beta distribution can be used to model events that are constrained to take place within an interval defined by a minimum and maximum value (in this case 0/1 none vs. readmission). It represents all the possible values of a probability when we are uncertain of the precise probability. The frequency of the
Please cite this article as: Byrne D, et al, The dynamics of the emergency medical readmission — The underlying fundamentals, Eur J Intern Med (2017), https://doi.org/10.1016/j.ejim.2017.09.028
D. Byrne et al. / European Journal of Internal Medicine xxx (2017) xxx–xxx
occurrence and their probabilities are determined by observation. The alpha 1 and 2 shape parameters were 0.70 and 2.17 respectively. The calculated time to readmission for the 5th, 10th, 25th, 50th, 75th, and 90th centiles were 6, 14, 48, 145, 288 and 441 weeks; early readmissions (b28 days) represented 20.1% (95% CI: 19.8, 20.5) of all readmissions. An alternative good model fit was provided by the Burr distribution; the latter is a unimodal family of distributions with a wide variety of shapes used to model a wide variety of phenomena including household income, risk (insurance) and travel time. It appears that these are flexible devices for modeling complex real life or medical event rates or uncertainty where there is limited information on the underlying basis for the event rate. As such they do not provide much insight into the dynamic of the process, the frequency, timing and probabilities being determined by empirical observations. Of course models may have utility as in hospitals readmissions reduction programs, with hospital performance assessed using risk adjustment measures — based on hierarchical logistic regression models [30]. Population based prediction models have also been investigated [31]. Interventions designed to prevent readmissions will inevitably have to contend with the specific ability to predict the high-risk patient, or groups likely to suffer a hospital readmission. The challenge, of course, is to accurately define groups of patients where one can intervene and offset a readmission without the patient suffering harm as a result. The importance of getting this right is illustrated by the fact that up to 29% of patients admitted to hospital die within a year of their index admission [32]. In that regard the study by Dharmarajan et al. [33] examined whether readmission reduction measures encouraged by the affordable care act had any adverse effect on mortality outcome. Over 6.5 million episodes of care for heart failure, myocardial infarction and pneumonia were reviewed. The reductions in 30 day risk-adjusted-readmission-rate (RARR) over the period were 0.053%, 0.044% and 0.033% respectively. There was no increase in 30-day risk adjusted mortality; however, the reduction in readmission rate was clearly small. The risk of readmissions is not just confined to serious life threatening conditions. An interesting study by Walsh et al. [34] looked at the outcomes of patients who had an index episode of care for a ‘poorly defined condition’. These admissions accounted for 21% of activity and whilst this group has a much lower mortality risk over the 12 months of followup they were as likely to be readmitted as a patient with a clearly defined diagnosis. In our study the AUROC of the multi-variable logistic model was 0.61 indicating an inability to predict usefully based on the selected factors; others have found similarly unimpressive ability to predict readmissions [12]. Of course the factors such as Acute Illness Severity Score [18,19], Charlson Co-Morbidity Index [20] and Chronic Disabling Score [21] and Deprivation Status [22,23] may be recognised, but this does not usefully translate into specifics regarding individual patients or timing. Although the logistic model could correctly predict readmissions with a rate of 72%, 28% of those with a positive classification were incorrect and 42% of with a negative classification were incorrect. Our study site is located in Dublin's inner city with a catchment population composed of high rates of social disadvantage when compared with other acute hospitals within the Dublin metropolitan area [35]. It is clear that hospitals located in areas of higher socio-economic disadvantage are at a potential disadvantage and this might explain our higher readmission rate of 9.6% compared with the UK average of 7% [13]. Irrespective, the current study indicates that readmission rates have not altered despite significant reductions in mortality rates. Our study has several limitations. Firstly, it was performed in a single-centre and, as previously referenced, serves a catchment area with rather high rates of socio-economic disadvantage. Therefore, our results may not be generalisable to other healthcare institutions and need to be interpreted with caution in this regard. Secondly, as the total numbers of patients in the least deprived groups (Q1, Q2) are significantly lower than in the more deprived groups (Q4, Q5), one cannot exclude entirely the possibility that patients from least deprived geographical
5
areas have attended other healthcare institutions, and therefore not all emergency hospital admissions in this social group are captured in this study. Nevertheless, our results are from a large dataset of over 96,000 hospital episodes over a 15-year period in the largest acute hospital serving Dublin's south inner city. 5. Conclusion We have shown that early readmission rates have not materially changed during a 15 year period where in-patient mortality has been significantly reduced. In addition, the kinetics of readmission may reflect the interaction of the underlying trajectory of chronic disease and social factors. Our ability to predict those who are likely to be readmitted is limited. Taken in conjunction with the experience that readmission avoidance strategies have mixed efficacy, and the fact that any hospital admission is associated a high 12 month mortality risk, there is clearly a requirement for a more systematic view of the coherence and effectiveness of using readmission avoidance as a key performance indicator. Rather, we would argue that the totality of clinical care, across the primary and secondary care spectrum, be optimized to ensure a patient centred experience that delivers the best outcome possible. LOS: length of stay, MDC: major disease category. Note patients with LOS N 30 days are not included. References [1] Benbassat J, Taragin M. Hospital readmissions as a measure of quality of health care: advantages and limitations. Arch Intern Med 2000;160(8):1074–81. [2] Moloney ED, Bennett K, Silke B. Patient and disease profile of emergency medical readmissions to an Irish teaching hospital. Postgrad Med J 2004;80:470–4. [3] Luthi. Is readmission to hospital an indicator of poor process of care for patients with heart failure? Qual Saf Health Care 2004;13. [4] Ashton CM, Del Junco DJ, Souchek J, Wray NP, Mansyur CL. The association between the quality of inpatient care and early readmission: a meta-analysis of the evidence. Med Care 1997;35(10):1044–59. [5] DesHarnais S, McMahon Jr LF, Wroblewski R. Measuring outcomes of hospital care using multiple risk-adjusted indexes. Health Serv Res 1991;26(4):425–45. [6] Clarke A. Readmission to hospital: a measure of quality or outcome? Qual Saf Health Care 2004;13(1):10. [7] Zook CJ, Moore FD. High-cost users of medical care. N Engl J Med 1980;302(18): 996–1002. [8] Anderson MA, Tyler D, Helms LB, Hanson KS, Sparbel KJH. Hospital readmission from a transitional care unit. J Nurs Care Qual 2005;20(1):26–35. [9] Chambers M, Clarke A. Measuring readmission rates. Br Med J 1990;301(6761): 1134–6. [10] Tierney AJ, Worth A. Review. Readmission of elderly patients to hospital. Age Ageing 1995;24(2):163–6. [11] Victor C, Jeffries S. Are re-admission rates a useful indication of outcome? Geriatric Med 1990:19–20. [12] Gorman J, Vellinga A, Gilmartin JJ, O'Keeffe ST. Frequency and risk factors associated with emergency medical readmissions in Galway University Hospitals. Ir J Med Sci 2010;179(2):255–8. [13] Blunt I, Bardsley M, Grove A, Clarke A. Classifying emergency 30-day readmissions in England using routine hospital data 2004–2010: what is the scope for reduction? EMJ: Emerg Med J; 2014. [14] van Walraven C, Jennings A, Forster AJ. A meta-analysis of hospital 30-day avoidable readmission rates. J Eval Clin Pract 2012;18(6):1211–8. [15] Conway R, O'Riordan D, Silke B. Long-term outcome of an AMAU—a decade's experience. Q J Med 2014;107(1):43–9. [16] O'Loughlin R, Allwright S, Barry J, Kelly A, Teljeur C. Using HIPE data as a research and planning tool: limitations and opportunities. Ir J Med Sci 2005;174(2):40–5 [discussion 52-7]. [17] Mikulich O, Callaly E, Bennett K, Silke B, O' Riordan D. The increased mortality associated with a weekend emergency admission is due to increased illness severity and altered case-mix. Acute Med 2011;10(4):181–6. [18] Silke B, Kellett J, Rooney T, Bennett K, O'Riordan D. An improved medical admissions risk system using multivariable fractional polynomial logistic regression modelling. Q J Med 2010;103(1):23–32. [19] O'Sullivan E, Callely E, O'Riordan D, Bennett K, Silke B. Predicting outcomes in emergency medical admissions — role of laboratory data and co-morbidity. Acute Med 2012;2:59–65. [20] Charlson ME, Pompei P, Ales KL, MacKenzie CR. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J Chronic Dis 1987;40(5):373–83. [21] Chotirmall SH, Picardo S, Lyons J, Alton M, O'Riordan D, Silke B. Disabling disease codes predict worse outcomes for acute medical admissions. Intern Med J 2014 [n/ a-n/a].
Please cite this article as: Byrne D, et al, The dynamics of the emergency medical readmission — The underlying fundamentals, Eur J Intern Med (2017), https://doi.org/10.1016/j.ejim.2017.09.028
6
D. Byrne et al. / European Journal of Internal Medicine xxx (2017) xxx–xxx
[22] Kelly A, Teljeur C. SAHRU National Deprivation Index. Dublin: Trinity College; 2007Available from: http://www.sahru.tcd.ie/services/deprivation/DeprivationFiles/ WebReport07.pdf. [23] Conway R, Galvin S, Coveney S, O'Riordan D, Silke B. Deprivation as an outcome determinant in emergency medical admissions. QJM 2013;106(3):245–51. [24] Shirmat M. Algorithm 112: position of point relative to polygon. ACM Comm 1962; 5:434. [25] Chotirmall SH, Callaly E, Lyons J, O'Connell B, Kelleher M, Byrne D, et al. Blood cultures in emergency medical admissions: a key patient cohort. Eur J Emerg Med 2014. [26] Department of Health. Payment by results guidance for 2012–13. Gateway reference 17250. London: Department of Health; 2012. [27] Readmissions Reduction Program. Available from: http://cms.gov/Medicare/ Medicare-Fee-for-Service-Payment/AcuteInpatientPPS/Readmissions-ReductionProgram.html; 2013. [28] Clarke A. Are readmissions avoidable? BMJ 1990;301(6761):1136–8. [29] Rose G. Sick individuals and sick populations. Int J Epidemiol 2001;30(3):427–32.
[30] McIlvennan CK, Eapen ZJ, Allen LA. Hospital readmissions reduction program. Circulation 2015;131(20):1796–803. [31] Helm JE, Alaeddini A, Stauffer JM, Bretthauer KM, Skolarus TA. Reducing hospital readmissions by integrating empirical prediction with resource optimization. Prod Oper Manag 2015;25(2):233–57. [32] Clark D, Armstrong M, Allan A, Graham F, Carnon A, Isles C. Imminence of death among hospital inpatients: prevalent cohort study. Palliat Med 2014;28(6):474–9. [33] Dharmarajan K, Wang Y, Lin Z, Normand ST, Ross JS, Horwitz LI, et al. Association of changing hospital readmission rates with mortality rates after hospital discharge. JAMA 2017;318(3):270–8. [34] Walsh B, Roberts HC, Nicholls PG. Features and outcomes of unplanned hospital admissions of older people due to ill-defined (R-coded) conditions: retrospective analysis of hospital admissions data in England. BMC Geriatr 2011;11:62. [35] Walsh JB, Coakley D, Murphy C, Coakley JD, Boyle E, Johnson H. Demographic profile of the elderly population in Dublin accident and emergency hospital catchment areas. Ir Med J 2004;97(3):84–6.
Please cite this article as: Byrne D, et al, The dynamics of the emergency medical readmission — The underlying fundamentals, Eur J Intern Med (2017), https://doi.org/10.1016/j.ejim.2017.09.028
Contents lists available at ScienceDirect
European Journal of Internal Medicine journal homepage: www.elsevier.com/locate/ejim
Reflections in Internal Medicine
The dynamics of the emergency medical readmission — The underlying fundamentals Declan Byrne, Deirdre O'Riordan, Richard Conway, Sean Cournane, Bernard Silke ⁎ Department of Internal Medicine, St James's Hospital, Dublin 8, Ireland
a r t i c l e
i n f o
Article history: Received 5 September 2017 Received in revised form 18 September 2017 Accepted 23 September 2017 Available online xxxx Keywords: Emergency medical admission Readmissions Dynamic patterns
a b s t r a c t Background: Hospital readmissions are a perennial problem. We reviewed readmissions to one institution (2002– 2015) and investigated their dynamics. Methods: 96,474 emergency admissions (in 50,701 patients) to an Irish hospital over a 15-year period were studied, and patterns surrounding early (b 28 days) and late (any other) readmissions determined. Univariate and logistic or truncated Poisson regression methods were employed. Results: Early readmission rate averaged 9.6% (95% CI: 9.4, 9.8) with a low/high of 8.4% (95% CI: 7.8, 9.1) and 10.3% (95% CI: 9.6, 11.0) respectively with no overall time trend. Early readmissions represented 20.1% (95% CI: 19.8, 20.5) of emergency medical readmissions. Median time to first readmission was 55 weeks (95% CI: 13, 159), time to second was 35 weeks (95% CI: 9, 98); by the 7th/8th readmissions, intervals were 13 weeks (95% CI: 4, 36) and 11 weeks (95% CI: 4, 30). Readmissions were older 67.1 years (95% CI: 48.3, 79.2) vs. single admissions 53.9 years (34.3, 72.4) and stayed longer — 5.8 days (2.7, 10.6) vs. 3.9 days (1.5, 8.0). Readmissions had more Acute Illness Severity, Charlson Co-Morbidity and Chronic Disabling Disease. Between 2002 and 2015 the logistic adjusted model of 30-day in-hospital mortality reduced from 6.1% (95% CI: 5.7, 6.5) to 4.4% (95% CI: 4.1, 4.7) (RRR 30.4%). Conclusion: Early hospital readmission rate did not change over 15 years despite improvements in hospital mortality outcomes. Readmissions have a consistent pattern related to patient illness and social characteristics; the fundamentals are driven by disease progression over time. © 2017 European Federation of Internal Medicine. Published by Elsevier B.V. All rights reserved.
1. Introduction There is a rising trend of unplanned hospital medical readmissions; a particular concern are those occurring early — defined as within 1 month of hospital discharge [1]. This trend is of concern due to the questions it raises regarding the quality of care during the index hospitalisation, and also because of the overall increased burden on the acute hospital service provider [1,2]. There is the view that an early hospital readmission reflects inferior hospital care; a meta-analysis of 16 studies [3] suggested that the risk of early readmission (within 31 days) was increased by 55% with care that was judged to be of relatively low quality. Further a systematic review by Ashton et al. indicated that on average, substandard care increased the risk of early readmission by 24% [4]. However, when DesHarnais et al. [5] ranked 300 hospitals on 3 risk adjusted indices of hospital quality, mortality, readmissions and complications, there was no relationship between a hospital's ranking on any one of these indices and it's ranking on the other two. On balance the evidence suggests that readmission rates, uncorrected for confounding medical, social and hospital factors, are a poor guide to quality of care [1,6]. ⁎ Corresponding author. E-mail address: [email protected] (B. Silke).
It has been estimated that 13% of inpatients in the United States use more than half of all hospital resources through repeated admissions [7, 8]. Reported rates of unplanned emergency readmissions are 15.1% at 28 days from North East Thames [9], 28% at three months in Edinburgh [10], 38% at six months in London [11], and 19.5% at one year in Galway [12]. Overall, it has been estimated that 7% of hospital discharges result in a readmission [13], but b25% of these may be preventable [14]. The objective of this paper is to evaluate the evidence base as to whether interventions aiming to address preventing early readmissions are realistically based; we review the dynamics of readmissions with a prospectively collected database of nearly 100,000 emergency medical admissions, between 2002 and 2016, and the evidence of the data.
2. Methods 2.1. Background St James's Hospital (SJH) serves as a secondary care centre for emergency admissions for its catchment area of 270,000 adults. Emergency medical patients are admitted from the Emergency Department to an Acute Medical Admission Unit (AMAU) — the operation and outcome of which have been described elsewhere [15].
https://doi.org/10.1016/j.ejim.2017.09.028 0953-6205/© 2017 European Federation of Internal Medicine. Published by Elsevier B.V. All rights reserved.
Please cite this article as: Byrne D, et al, The dynamics of the emergency medical readmission — The underlying fundamentals, Eur J Intern Med (2017), https://doi.org/10.1016/j.ejim.2017.09.028
2
D. Byrne et al. / European Journal of Internal Medicine xxx (2017) xxx–xxx
2.2. Data collection
2.5. Statistical methods
For audit purposes we employed an anonymous patient database assembling core information about each clinical episode from elements contained on the patient administration system, the national hospital in-patient enquiry (HIPE) scheme [16], the patient electronic record, and laboratory system. Data captured includes the unique hospital number, patient demographics, principal and secondary diagnoses, procedures performed, admission and discharge dates. Additional information cross-linked and automatically uploaded to the database includes physiological and laboratory parameters. From modeling laboratory data collected at time of hospital admission we developed a predictive algorithm based on serum sodium, potassium, urea, albumin, red cell distribution width, and white blood cell count. The underlying principle is that deviation beyond the boundaries of ‘normal homeostasis’ is an estimate of risk, although the relationship is non-linear and differs for each variable. Six groups were originally defined with a 30-day mortality risk increasing in an exponential fashion [17]. We assessed the ability of known predictors — Acute Illness Severity [18,19], Charlson Co-Morbidity Index [20] and Chronic Disabling Score [21] to predict any end-points, including 30day hospital mortality, Length of Hospital stay (LOS) and any readmission (early b 28 days or any).
Descriptive statistics were calculated for background demographic data, including means/standard deviations (SD), medians/inter-quartile ranges (IQR), or percentages. Comparisons between categorical variables and mortality were made using chi-square tests. Logistic regression analysis was used to examine all significant outcome predictors (p b 0.10 from the univariate analysis) on 30-day in hospital mortality; in addition to the Acute Illness Severity Score [18, 19], Charlson Co-Morbidity Index [20] and Chronic Disabling Score [21] to predict any end-points, including 30-day hospital mortality, Length of Hospital stay (LOS) and any readmission (early b28 days or any). We then used backwards and forwards stepwise methods to determine the optimal predictors, while testing the goodness-of-fit using the Hosmer and Lemeshow's tests. The latter test is based on the expectation, that the predicted and observed frequency should match closely, and that the more closely they match, the better the fit. Adjusted Odds Ratios (OR), Relative Risk Reduction (RRR), and Number Needed to Treat (NNT) and 95% confidence intervals (CI) were calculated where appropriate. Statistical significance at P b 0.05 was assumed throughout. Stata v.13.1 (Stata Corporation, College Station, Texas, USA) was used for analysis. 3. Results
2.3. Deprivation status and calculations 3.1. Patient demographics A National Deprivation Index for Health Services Research was derived by the Small Area Health Research Unit (SAHRU) at Trinity College Dublin [22]. The census data from the Central Statistics Office (CSO) (1991, was utilized to compute a functional index of deprivation using the 3440 electoral divisions (ED's) within the Republic of Ireland; these are smallest administrative areas for which population statistics are released. A weighted combination of four indicators, relating to unemployment, social class, type of housing tenure and car ownership ranked each small area [22]. The national deprivation index scores were ranked by decile from low (least deprived) to high (most deprived) according to their ranked raw scores [23]. There are 74 EDs in the hospital catchment area with a population of 210,443 persons, as per 2006. The median population per ED was 2845 (IQR 2020, 3399). These areas were ranked nationally as Deprivation Quintile I (n = 14), Quintile III (n = 6), Quintile IV (n = 5) and Quintile V (n = 49). Using an address, data at an individual level were geo-coded and matched with the SAHRU deprivation raw score and related rank quintile. These attribute data were joined to the small area polygon geometries based upon their relative geographic positions, using the ArcGIS 10 Geographic Information System software implementation of the Pointin-Polygon algorithm, as outlined by Shimrat [24].
There were a total of 96,306 episodes recorded in 50,701 patients admitted as medical emergencies between 2002 and 2016. The proportion of males was 48.9%. The median (IQR) length of stay (LOS) was 4.7 (1.9, 9.1) days. The median (IQR) age was 58.7 (37.4, 76.0) years, with the upper 10% boundary at 86.1 yr. The major disease categories (MDC) were respiratory (25.4%), cardiovascular (16.6%), neurological (16.3%), gastrointestinal (10.2%), hepatobiliary (4.5%) and renal (4.9%). The majority of admissions were from the more socially deprived section of society (Quintiles IV & V: 89.7%). Table 1 describes the demographic characteristic of the first and subsequent readmissions. Readmissions were older 67.1 years (95% CI: 48.3, 79.2) vs. single admissions 53.9 years (34.3, 72.4) and stayed longer — readmission 5.8 days (2.7, 10.6) vs. single admission 3.9 days (1.5, 8.0). Readmissions had more Acute Illness Severity, Charlson Co-Morbidity and Chronic Disabling Disease (Table 1). Readmissions had more respiratory, but less cardiovascular and neurological MDC codings. Total readmissions (i.e. not just early b 28 days) were clearly time dependent; their cumulative proportion increased from 23.5% in 2003 to 45.9% in 2016. Overall early readmissions (b28 days) represented 9.6% (95% CI: 9.4, 9.8) of emergency medical admitted episodes. 3.2. Time pattern of readmission (Figs. 1 & 2)
2.4. Study inclusion criteria For this study, data was related to all emergency general medical patients admitted to SJH between 2002 and 2016. Each emergency medical patient was referred to the team of the ‘on-call’ Acute Medicine Consultant – ‘on-take’ for a 24 h period – most ~ 90% remained under the care of the admitting consultant for the duration of their admission. Approximately 9.9% of our patients stay N30 days with a median LOS of 54.8 days (IQR 38.8, 97.2). Consequently, the LOS data represents a highly skewed distribution. Although the clinical episode is complete for the majority by day 30, some patients remain for social reasons related to the lack of long-term care facilities. We have therefore chosen a truncated end-point (at the 30-day endpoint) for mortality analyses, to avoid these additional confounders. Readmissions obviously consider all patient episodes, irrespectively of length of hospital stay.
Between 2002 and 2016, from a total of 96,306 episodes there were 33,254 episodes with no readmission (34.5%); the readmission rate overall (excluding the first admission of a series of readmissions) was 47.6% (95% CI: 47.3, 48.0). The respective calculated readmission rates at 1, 3, 5 and 10 yr were 32.4%, 40.6%, 48.9% and 53.3% respectively. Over the 15 yr period, early readmissions (b 28 days) represented 20.1% (95% CI: 19.8, 20.5) of all readmissions. Following an emergency medical admission, the risk of a readmission is high initially but rapidly falls albeit with a long tail (Fig. 1). The best fit of the time to readmission was the beta density distribution function. The alpha 1 and 2 parameters were 0.70 and 2.17 respectively. The calculated time to readmission for the 5th, 10th, 25th, 50th, 75th, and 90th centiles were 6, 14, 48, 145, 288 and 441 weeks. The time from the first to each subsequent readmission is described (Fig. 2). The median time was calculated in weeks between the first and
Please cite this article as: Byrne D, et al, The dynamics of the emergency medical readmission — The underlying fundamentals, Eur J Intern Med (2017), https://doi.org/10.1016/j.ejim.2017.09.028
D. Byrne et al. / European Journal of Internal Medicine xxx (2017) xxx–xxx
3
Table 1 Characteristics of emergency medical admissions (2002–2016). Factor N Gender Outcome Age, median (IQR) Length of stay (days) Acute illness severity
Charlson Index
Disabling disease
Sepsis status
MDC respiratory MDC cardiovascular MDC neurology
Level Male Female Alive Died
1 2 3 4 5 6 0 1 2 0 1 2 3 4 0 1 2 0 1 0 1 0 1
No readmission
1+ readmission
30,236 14,725 (48.7%) 15,511 (51.3%) 28,328 (93.7%) 1908 (6.3%) 53.9 (34.3, 72.4) 3.9 (1.5, 8.0) 1705 (6.7%) 3395 (13.2%) 4475 (17.5%) 4738 (18.5%) 4155 (16.2%) 7168 (28.0%) 18,309 (60.6%) 6205 (20.5%) 5722 (18.9%) 5272 (17.4%) 9835 (32.5%) 8171 (27.0%) 4680 (15.5%) 2278 (7.5%) 23,129 (76.5%) 6059 (20.0%) 1048 (3.5%) 24,570 (81.3%) 5666 (18.7%) 25,116 (83.1%) 5120 (16.9%) 24,043 (79.5%) 6193 (20.5%)
56,347 27,608 (49.0%) 28,739 (51.0%) 54,287 (96.3%) 2060 (3.7%) 67.1 (48.3, 79.2) 5.8 (2.7, 10.6) 845 (1.5%) 2595 (4.6%) 5697 (10.1%) 9025 (16.0%) 12,070(21.4%) 26,115 (46.3%) 21,441 (38.2%) 17,677 (31.5%) 17,052 (30.4%) 4223 (7.5%) 11,956 (21.2%) 17,234 (30.6%) 13,627 (24.2%) 9307 (16.5%) 42,865 (76.1%) 11,517 (20.4%) 1965 (3.5%) 40,033 (71.0%) 16,314 (29.0%) 47,108 (83.6%) 9239 (16.4%) 48,447 (86.0%) 7900 (14.0%)
P-value 0.41 b0.001 b0.001 b0.001 b0.001
b0.001
b0.001 Fig. 2. The data illustrate the cumulative time (weeks) between consecutive readmissions. The interval between each readmission progressively declined and approximated to an exponential model. 0.36
3.3. Comparison of the mortality and readmissions rates over time (Fig. 3) b0.001 0.043 b0.001
each subsequent admission. A mathematical model, a one-phase exponential association, satisfactorily fits the cumulative time intervals in weeks of the first eight readmissions, with a correlation coefficient of 0.99. As time advances the duration between consecutive admissions progressively shortens. The median time to the first readmission was the longest at 55 weeks (95% CI: 13, 159), the time to the second was less at 35 weeks (95% CI: 9, 98) but by the 7th and 8th readmissions, the corresponding intervals were 13 weeks (95% CI: 4, 36) and 11 weeks (95% CI: 4, 30).
The St James' Hospital medical emergency database was set up in 2002, primarily to track the impact of a new initiative in acute medicine, namely the acute medical admission unit (AMAU). While there may be factors between 2002 and 2016 that may have improved outcome over those years, it is of interest to look at mortality and readmission rates over these years. The change in 30-day in-hospital mortality over the study period is shown in Fig. 3. Calculated by episode (ignoring readmissions) the logistic adjusted model of 30-day in-hospital mortality showed a decline from 6.1% (95% CI: 5.7, 6.5) in 2002 to 4.4% (95% CI: 4.1, 4.7); the relative risk reduction was 30.4% with a NNT of 60.1. Calculating on unique patients (one episode only counted — last if N one admission), the logistic adjusted model of 30-day in-hospital mortality showed a decline from 17.3% (95% CI: 16.4, 18.2) in 2002 to 8.0% (95% CI: 7.5, 8.5); the relative risk reduction was 61% with a NNT of 13.5. These remarkable improvements in outcome have to be compared with the readmissions over time. On average the early readmission rate (b28 days) averaged 9.6% (95% CI: 9.4, 9.8) with a low and high of 8.4 (95% CI: 7.8, 9.1) and 10.3 (95% CI: 9.6, 11.0) respectively with no overall time trend.
3.4. Importance of deprivation status and illness severity on the hospital readmission rate (Fig. 4)
Fig. 1. Time to readmission following an index emergency medical admission. The model best fit was described by the Beta density distribution with parameters alpha1 and alpha2 of 0.70 and 2.17 respectively. Time (weeks) to readmission at the 75, 50, 25 and 10th centiles was 50, 145, 288 and 432 weeks respectively.
The data clearly showed that readmissions to our acute hospital were strongly related to the Deprivation status of the patients. Over the 15 year period, we calculated the total number of readmissions from each small area (i.e. Electoral Division); relating these to each ED population allowed a readmission rate to be calculated (readmission rate / 1000 population) which in turn allowed us to calculate the incidence rate ratio (IRR). Readmission was much more likely from a high deprivation area — IRR 1.69 (95% CI: 1.45, 1.98). An eight fold (×8) variation in calculated readmission rate was identified between the least deprived group (Q1) — 2.5/1000 (95% CI: 1.24, 3.76), and the most deprived group (Q5) — 20.6/1000 (95% CI: 15.6, 25.6. There were other predictive factors assessed in the multiple variable logistic model. These included the Acute Illness Severity (18, 19); those with higher illness scores were more likely have a readmission — IRR 1.37 (95% CI: 1.27, 1.48). Age might be conjectured as a likely association with an increased readmission rate; however adjusted for illness severity and complexity this was not the case IRR 0.98 (95% CI: 0.98,
Please cite this article as: Byrne D, et al, The dynamics of the emergency medical readmission — The underlying fundamentals, Eur J Intern Med (2017), https://doi.org/10.1016/j.ejim.2017.09.028
4
D. Byrne et al. / European Journal of Internal Medicine xxx (2017) xxx–xxx
Fig. 3. Between 2002 and 2016, the logistic adjusted model of 30-day in-hospital mortality showed a per episode decline from 6.1% (95% CI: 5.7, 6.5) to 4.4% (95% CI: 4.1, 4.7) — a relative risk reduction of 30.4%. The logistic adjusted model for the early readmission rate (b28 days) averaged 9.6% (95% CI: 9.4, 9.8) of episodes with a low and high of 8.4 and 10.9 respectively with no overall time trend.
0.99). Complexity could be inferred from either the Chronic Disabling Disease Score [21] or the Charlson Index [20]. A higher Chronic Disabling Score [21] was strongly associated with more readmissions — IRR 1.20 (95% CI: 1.14, 1.26); higher Charlson scores predicted a lower readmission likelihood — IRR 0.87 (95% CI: 0.81, 0.93). The Charlson Index is weighted with cancer and cardiovascular bias — which are likely to have a worse prognosis; similarly the presence of sepsis [25], for possibly similar reasons, was less likely to be associated with further readmissions — IRR 0.81 (95% CI: 0.75, 0.87). The AUROC in the model was 0.61 (95%CI: 0.60, 0.62).
Fig. 4. The hospital readmission incidence (rate/1000 population) was strongly influenced by the deprivation status. The Irish National Index ranked the catchment small areas from low Q1 to high Q5 deprivation. The predicted probabilities were derived from the multiple variable truncated Poisson model; the effect with confidence interval is plotted based on the latter prediction (No Q2 Electoral Divisions exist within the hospital catchment area).
4. Discussion Emergency admissions and readmissions are rising in many countries [13]; healthcare organisations such as the Department of Health in England and Medicare and Medicaid in the USA have produced guidance on restricting payments for readmission within 30 days of discharge from a previous (index) admission [26,27]. The crucial question that follows is, of course, as to what is the extent to which readmissions are preventable? [28]. Our data, collected in one institution over 15 years, demonstrates that following an emergency medical admission, the risk of a readmission is high initially but rapidly declined with a long tail. On average the early readmission rate (b28 days) averaged 9.6% (95% CI: 9.4, 9.8) with a low and high of 8.4% (95% CI: 7.8, 9.1) and 10.3% (95% CI: 9.6, 11.0) respectively with no overall time trend. Blunt et al. [13] estimated that early readmissions over a 6 year period were equivalent to 7.0% of hospital discharges. However, early readmissions are part of a distribution continuum; the question not being addressed is whether early readmissions can be disassociated from late readmissions. Or more specifically whether it is realistic to intervene and attempt to selectively alter part of a distribution? Geoffrey Rose focused on this issue in a classic paper [29]; he argued in favour of the population approach (targeting the entire distribution) rather than some specific component. Or as he aptly put it, the distribution is the way it is for complex reasons that one may not comprehend and it would resist well-meaning enthusiastic interventions designed to modify one aspect of the curve. Modeling the curve, suggested that the best fit was with a beta distribution model. The beta distribution is a family of continuous probability distributions defined on the interval [0, 1] parameterized by two positive shape parameters, typically denoted by α and β. The latter control the shape of the distribution. The Beta distribution can be used to model events that are constrained to take place within an interval defined by a minimum and maximum value (in this case 0/1 none vs. readmission). It represents all the possible values of a probability when we are uncertain of the precise probability. The frequency of the
Please cite this article as: Byrne D, et al, The dynamics of the emergency medical readmission — The underlying fundamentals, Eur J Intern Med (2017), https://doi.org/10.1016/j.ejim.2017.09.028
D. Byrne et al. / European Journal of Internal Medicine xxx (2017) xxx–xxx
occurrence and their probabilities are determined by observation. The alpha 1 and 2 shape parameters were 0.70 and 2.17 respectively. The calculated time to readmission for the 5th, 10th, 25th, 50th, 75th, and 90th centiles were 6, 14, 48, 145, 288 and 441 weeks; early readmissions (b28 days) represented 20.1% (95% CI: 19.8, 20.5) of all readmissions. An alternative good model fit was provided by the Burr distribution; the latter is a unimodal family of distributions with a wide variety of shapes used to model a wide variety of phenomena including household income, risk (insurance) and travel time. It appears that these are flexible devices for modeling complex real life or medical event rates or uncertainty where there is limited information on the underlying basis for the event rate. As such they do not provide much insight into the dynamic of the process, the frequency, timing and probabilities being determined by empirical observations. Of course models may have utility as in hospitals readmissions reduction programs, with hospital performance assessed using risk adjustment measures — based on hierarchical logistic regression models [30]. Population based prediction models have also been investigated [31]. Interventions designed to prevent readmissions will inevitably have to contend with the specific ability to predict the high-risk patient, or groups likely to suffer a hospital readmission. The challenge, of course, is to accurately define groups of patients where one can intervene and offset a readmission without the patient suffering harm as a result. The importance of getting this right is illustrated by the fact that up to 29% of patients admitted to hospital die within a year of their index admission [32]. In that regard the study by Dharmarajan et al. [33] examined whether readmission reduction measures encouraged by the affordable care act had any adverse effect on mortality outcome. Over 6.5 million episodes of care for heart failure, myocardial infarction and pneumonia were reviewed. The reductions in 30 day risk-adjusted-readmission-rate (RARR) over the period were 0.053%, 0.044% and 0.033% respectively. There was no increase in 30-day risk adjusted mortality; however, the reduction in readmission rate was clearly small. The risk of readmissions is not just confined to serious life threatening conditions. An interesting study by Walsh et al. [34] looked at the outcomes of patients who had an index episode of care for a ‘poorly defined condition’. These admissions accounted for 21% of activity and whilst this group has a much lower mortality risk over the 12 months of followup they were as likely to be readmitted as a patient with a clearly defined diagnosis. In our study the AUROC of the multi-variable logistic model was 0.61 indicating an inability to predict usefully based on the selected factors; others have found similarly unimpressive ability to predict readmissions [12]. Of course the factors such as Acute Illness Severity Score [18,19], Charlson Co-Morbidity Index [20] and Chronic Disabling Score [21] and Deprivation Status [22,23] may be recognised, but this does not usefully translate into specifics regarding individual patients or timing. Although the logistic model could correctly predict readmissions with a rate of 72%, 28% of those with a positive classification were incorrect and 42% of with a negative classification were incorrect. Our study site is located in Dublin's inner city with a catchment population composed of high rates of social disadvantage when compared with other acute hospitals within the Dublin metropolitan area [35]. It is clear that hospitals located in areas of higher socio-economic disadvantage are at a potential disadvantage and this might explain our higher readmission rate of 9.6% compared with the UK average of 7% [13]. Irrespective, the current study indicates that readmission rates have not altered despite significant reductions in mortality rates. Our study has several limitations. Firstly, it was performed in a single-centre and, as previously referenced, serves a catchment area with rather high rates of socio-economic disadvantage. Therefore, our results may not be generalisable to other healthcare institutions and need to be interpreted with caution in this regard. Secondly, as the total numbers of patients in the least deprived groups (Q1, Q2) are significantly lower than in the more deprived groups (Q4, Q5), one cannot exclude entirely the possibility that patients from least deprived geographical
5
areas have attended other healthcare institutions, and therefore not all emergency hospital admissions in this social group are captured in this study. Nevertheless, our results are from a large dataset of over 96,000 hospital episodes over a 15-year period in the largest acute hospital serving Dublin's south inner city. 5. Conclusion We have shown that early readmission rates have not materially changed during a 15 year period where in-patient mortality has been significantly reduced. In addition, the kinetics of readmission may reflect the interaction of the underlying trajectory of chronic disease and social factors. Our ability to predict those who are likely to be readmitted is limited. Taken in conjunction with the experience that readmission avoidance strategies have mixed efficacy, and the fact that any hospital admission is associated a high 12 month mortality risk, there is clearly a requirement for a more systematic view of the coherence and effectiveness of using readmission avoidance as a key performance indicator. Rather, we would argue that the totality of clinical care, across the primary and secondary care spectrum, be optimized to ensure a patient centred experience that delivers the best outcome possible. LOS: length of stay, MDC: major disease category. Note patients with LOS N 30 days are not included. References [1] Benbassat J, Taragin M. Hospital readmissions as a measure of quality of health care: advantages and limitations. Arch Intern Med 2000;160(8):1074–81. [2] Moloney ED, Bennett K, Silke B. Patient and disease profile of emergency medical readmissions to an Irish teaching hospital. Postgrad Med J 2004;80:470–4. [3] Luthi. Is readmission to hospital an indicator of poor process of care for patients with heart failure? Qual Saf Health Care 2004;13. [4] Ashton CM, Del Junco DJ, Souchek J, Wray NP, Mansyur CL. The association between the quality of inpatient care and early readmission: a meta-analysis of the evidence. Med Care 1997;35(10):1044–59. [5] DesHarnais S, McMahon Jr LF, Wroblewski R. Measuring outcomes of hospital care using multiple risk-adjusted indexes. Health Serv Res 1991;26(4):425–45. [6] Clarke A. Readmission to hospital: a measure of quality or outcome? Qual Saf Health Care 2004;13(1):10. [7] Zook CJ, Moore FD. High-cost users of medical care. N Engl J Med 1980;302(18): 996–1002. [8] Anderson MA, Tyler D, Helms LB, Hanson KS, Sparbel KJH. Hospital readmission from a transitional care unit. J Nurs Care Qual 2005;20(1):26–35. [9] Chambers M, Clarke A. Measuring readmission rates. Br Med J 1990;301(6761): 1134–6. [10] Tierney AJ, Worth A. Review. Readmission of elderly patients to hospital. Age Ageing 1995;24(2):163–6. [11] Victor C, Jeffries S. Are re-admission rates a useful indication of outcome? Geriatric Med 1990:19–20. [12] Gorman J, Vellinga A, Gilmartin JJ, O'Keeffe ST. Frequency and risk factors associated with emergency medical readmissions in Galway University Hospitals. Ir J Med Sci 2010;179(2):255–8. [13] Blunt I, Bardsley M, Grove A, Clarke A. Classifying emergency 30-day readmissions in England using routine hospital data 2004–2010: what is the scope for reduction? EMJ: Emerg Med J; 2014. [14] van Walraven C, Jennings A, Forster AJ. A meta-analysis of hospital 30-day avoidable readmission rates. J Eval Clin Pract 2012;18(6):1211–8. [15] Conway R, O'Riordan D, Silke B. Long-term outcome of an AMAU—a decade's experience. Q J Med 2014;107(1):43–9. [16] O'Loughlin R, Allwright S, Barry J, Kelly A, Teljeur C. Using HIPE data as a research and planning tool: limitations and opportunities. Ir J Med Sci 2005;174(2):40–5 [discussion 52-7]. [17] Mikulich O, Callaly E, Bennett K, Silke B, O' Riordan D. The increased mortality associated with a weekend emergency admission is due to increased illness severity and altered case-mix. Acute Med 2011;10(4):181–6. [18] Silke B, Kellett J, Rooney T, Bennett K, O'Riordan D. An improved medical admissions risk system using multivariable fractional polynomial logistic regression modelling. Q J Med 2010;103(1):23–32. [19] O'Sullivan E, Callely E, O'Riordan D, Bennett K, Silke B. Predicting outcomes in emergency medical admissions — role of laboratory data and co-morbidity. Acute Med 2012;2:59–65. [20] Charlson ME, Pompei P, Ales KL, MacKenzie CR. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J Chronic Dis 1987;40(5):373–83. [21] Chotirmall SH, Picardo S, Lyons J, Alton M, O'Riordan D, Silke B. Disabling disease codes predict worse outcomes for acute medical admissions. Intern Med J 2014 [n/ a-n/a].
Please cite this article as: Byrne D, et al, The dynamics of the emergency medical readmission — The underlying fundamentals, Eur J Intern Med (2017), https://doi.org/10.1016/j.ejim.2017.09.028
6
D. Byrne et al. / European Journal of Internal Medicine xxx (2017) xxx–xxx
[22] Kelly A, Teljeur C. SAHRU National Deprivation Index. Dublin: Trinity College; 2007Available from: http://www.sahru.tcd.ie/services/deprivation/DeprivationFiles/ WebReport07.pdf. [23] Conway R, Galvin S, Coveney S, O'Riordan D, Silke B. Deprivation as an outcome determinant in emergency medical admissions. QJM 2013;106(3):245–51. [24] Shirmat M. Algorithm 112: position of point relative to polygon. ACM Comm 1962; 5:434. [25] Chotirmall SH, Callaly E, Lyons J, O'Connell B, Kelleher M, Byrne D, et al. Blood cultures in emergency medical admissions: a key patient cohort. Eur J Emerg Med 2014. [26] Department of Health. Payment by results guidance for 2012–13. Gateway reference 17250. London: Department of Health; 2012. [27] Readmissions Reduction Program. Available from: http://cms.gov/Medicare/ Medicare-Fee-for-Service-Payment/AcuteInpatientPPS/Readmissions-ReductionProgram.html; 2013. [28] Clarke A. Are readmissions avoidable? BMJ 1990;301(6761):1136–8. [29] Rose G. Sick individuals and sick populations. Int J Epidemiol 2001;30(3):427–32.
[30] McIlvennan CK, Eapen ZJ, Allen LA. Hospital readmissions reduction program. Circulation 2015;131(20):1796–803. [31] Helm JE, Alaeddini A, Stauffer JM, Bretthauer KM, Skolarus TA. Reducing hospital readmissions by integrating empirical prediction with resource optimization. Prod Oper Manag 2015;25(2):233–57. [32] Clark D, Armstrong M, Allan A, Graham F, Carnon A, Isles C. Imminence of death among hospital inpatients: prevalent cohort study. Palliat Med 2014;28(6):474–9. [33] Dharmarajan K, Wang Y, Lin Z, Normand ST, Ross JS, Horwitz LI, et al. Association of changing hospital readmission rates with mortality rates after hospital discharge. JAMA 2017;318(3):270–8. [34] Walsh B, Roberts HC, Nicholls PG. Features and outcomes of unplanned hospital admissions of older people due to ill-defined (R-coded) conditions: retrospective analysis of hospital admissions data in England. BMC Geriatr 2011;11:62. [35] Walsh JB, Coakley D, Murphy C, Coakley JD, Boyle E, Johnson H. Demographic profile of the elderly population in Dublin accident and emergency hospital catchment areas. Ir Med J 2004;97(3):84–6.
Please cite this article as: Byrne D, et al, The dynamics of the emergency medical readmission — The underlying fundamentals, Eur J Intern Med (2017), https://doi.org/10.1016/j.ejim.2017.09.028