Factors predicting hospital length-of-stay after radical prostatectomy: a population-based study

The study setting was Ireland, which has a mixed public-private health care system. All residents are entitled to use the public health system. Public hospitals also offer private health care, and patients can opt to transfer from public to private care at any time. Patients can also choose to be treated in private hospitals. Thus, there are three categories of patients: (1) public patients treated within public hospitals; (2) private patients who pay for treatment within public hospitals, either out-of-pocket or through private health insurance; and (3) private patients who pay for treatment in private hospitals.

The primary data sources for this study were the National Cancer Registry (NCR) and the Hospital In-Patient Enquiry Scheme (HIPE). The NCR records demographic and clinical information for all cancers diagnosed in the population usually resident in Ireland [17]. Treatment received within the first year of diagnosis is also collected. Most (97.5%) registrations are made by tumour registration officers (TROs) who collate and abstract data from various sources - including pathology laboratories, radiotherapy clinics, patient administration systems, and medical records departments - according to internationally agreed guidelines. The remaining registrations are derived from death certificates (2%) and from general practitioners ( < 0.5%). Death certificates are provided by the Central Statistics Office [18]; date and cause of death in cases registered by the NCR are ascertained by linkage to death certificates using probabilistic matching methods. Completeness of case ascertainment is estimated to be in excess of 97% [19].

HIPE is a computer-based information system that collects data on discharges from all acute public hospitals in Ireland [20]. Limited demographic, clinical and administrative data are collected [21]. Data are subject to computer-based edits/checks at data entry and later validation checks [22]. The NCR is provided with all HIPE records which mention cancer or cancer-related procedures such as radiotherapy and chemotherapy. Private hospitals can volunteer to contribute data to HIPE, but coverage is incomplete so, in this study information from HIPE was limited to patients treated in public hospitals (as either public or private patients).

Figure 1 provides an overview of the creation of the analysis dataset. Prostate cancers (ICD10-O3: C61) newly diagnosed between 2002 and 2008 were identified from the NCR. Since RP is rarely conducted in older men [23], the dataset was then limited to men aged < 70 years at diagnosis who had a RP (ICD 9 CM procedure codes 60.3, 60.4, 60.5, 60.62) recorded by the NCR before the end of follow-up on the 31/12/2009. Using probabilistic matching techniques, these prostate cancers were linked to HIPE episodes. We excluded 94 men who had RP in a public hospital but had a previous diagnosis of cancer; 63 of these 94 men were diagnosed in the 12 months prior to the prostate cancer diagnosis. Almost all of the cancers (95%) that occurred in the year prior to the prostate cancer diagnosis were bladder cancer (C67).

Information on patient demographics, clinical and treatment data and provider volume were derived from NCR data. Deprivation was measured using an index created from 2002 census variables which is based on the patient’s area of residence at the time of diagnosis [24]. Information on tumour grade is routinely recorded as part of the cancer registration process using the WHO grading system. Information on Gleason score (GS) recorded in medical records was translated by TROs into WHO grading as follows: GS < 5 = grade 1; GS 5-7 = grade 2; and GS > 7 = grade 3. Stage was defined according to American Joint Committee on Cancer (AJCC) summary staging [25]. For each hospital the annual volume of RPs performed and the median for the entire period was determined. Hospitals were classified as “lower” or “higher” volume by ordering hospitals on the period median volume and splitting so that approximately 50% of patients treated fell into each category. The same approach was used for surgeon volume. Since surgeons can work in more than one hospital and in private and public hospitals, surgeon volume was calculated from NCR data based on all RPs they performed in either public or private hospitals during 2002–2008. “Higher-volume” was defined as >49 RPs per annum for hospitals and >17 RPs per annum for surgeons.

Information on LOS, co-morbidities, admission type (emergency/elective) and discharge status (i.e. whether the patient was “public” or “private” to the consultant at the time of discharge) was derived from HIPE. Each admission (and readmission) was categorised as emergency or elective according to relevant HIPE codes. A co-morbidity count for each patient, based on the Elixhauser index [26] - which was developed to predict a range of outcomes including length-of-stay - was derived from diagnoses recorded on HIPE during the index surgery episode; the prostate cancer diagnosis was disregarded in this calculation.

To determine LOS, HIPE episodes were ordered by date of admission and overlapping episodes combined. The date of first surgery recorded by the NCR was matched to the corresponding HIPE episode to identify the index surgery episode. LOS was defined as the number of days between the date of admission and the date of death or discharge, to either home or another non-acute care facility (e.g. nursing home). For the LOS analysis, cases were excluded if they had: no HIPE episode corresponding to the NCR treatment record; surgery in private hospitals; or surgery after the censoring date (31/12/2009) (Figure 1).

Duration of discharge was calculated as number of days from discharge following the RP to the date of next admission (if any). Patients whose discharge duration was ≤28 days were considered readmissions. This period was chosen for two reasons: firstly, because the 28-day emergency readmission rate is a key hospital performance indicator in some healthcare settings [27]; and secondly because catheters inserted routinely following RP would be expected to have been removed by this time.

Characteristics of patients who had RP and those who did not, and of those who had RP in public and private hospitals, were compared using chi-square tests. Median LOS was determined overall and by year. Trends in LOS by year were examined using the Kruskal-Wallis equality-of-populations rank test [28] and Cuzick’s non-parametric test for trend [29].

Since we are not aware of any national, or internationally, agreed definition of prolonged LOS following RP, in the primary analysis we defined a prolonged hospital stay as a stay of duration greater than the upper-quartile of LOS (>9 days). Multivariable logistic regression was used to identify factors which predicted prolonged LOS. Three types of variables were considered for inclusion in the model: socio-demographic (age, marital status, deprivation index, smoking status at diagnosis, patient status at discharge); clinical (grade, stage, co-morbidity); and care (hospital volume, surgeon volume, year of surgery). Four cases had missing information on marital status and were excluded entirely from the analysis dataset. Cases with missing data on other variables (deprivation index (n = 139), smoking status at diagnosis (n = 328), patient status at discharge (n = 112), grade (n = 43) and stage (n = 1128)) were included as a separate level in the multivariable analysis (see Additional file 1). Variables were included in the multivariable model if they were significant (p < 0.05) on likelihood ratio tests. Model goodness-of-fit was checked using the Hosmer and Lemeshow test [30].

We tested whether the effect of hospital volume on prolonged LOS was modified by surgeon volume by comparing two models: one including an interaction term between hospital and surgeon volume and one without this.

As the median (LOS = 8 days) and 75th percentile (LOS = 9 days) differed by only one day we conducted a sensitivity analyses where we repeated the main analysis using the 90th percentile to define prolonged LOS (>12 days). In a second sensitivity analysis we used aggregated LOS to define prolonged LOS. Aggregated LOS was the sum of LOS for the index RP episode plus all overnight readmissions occurring with 28 days of discharge from the index episode; in this analysis prolonged LOS was defined as the upper-quartile of aggregated LOS for all cases (>10 days).

Readmissions were tabulated by hospital and surgeon volume; these were further categorized into elective and emergency admissions. Reasons for readmission were obtained from HIPE diagnostic codes for the relevant episodes. All analyses were carried out using Stata 11 [31].

One of the major strengths of this study is that it is based on high-quality, population-based, cancer registration data (rather than hospital admission data alone), providing confidence that the patients included had prostate cancer. In addition, in our mixed public-private setting, where clinicians frequently work in both sectors, the availability of national cancer registry data made it possible to accurately quantify the entire (i.e. public and private) volume of cases for each surgeon.

Two thirds of men had RP in a public hospital and one third in a private hospital. Just over 4% of publicly-treated patients had no matching HIPE episode. A failure to find a match between cancer registrations and hospital episodes can occur for a number of reasons including: an inability to match records belonging to the same person due to typographical errors or missing information on either system, or no mention of cancer (or cancer-related procedures) on HIPE, in which case the episodes would not be available to the cancer registry. The missing episodes were distributed across hospitals and years.

We excluded 94 men who had another invasive cancer diagnosis (other than a non-melanoma skin cancer) prior to the prostate cancer diagnosis. The majority of these (67%) were diagnosed in the 12 months prior to the prostate cancer diagnosis. The diagnosis and treatment of recent previous cancers are likely to have an impact on the treatment decisions for the prostate cancer and hence may impact on LOS in men undergoing RP. This exclusion means that our results are generalizable to the population of men treated with RP for whom prostate cancer is their first cancer.

Our median LOS of 7 days in 2008 is slightly higher than reported in England (median 6 days, based on patients treated in the publically-funded NHS) [9]. Our median post-operative stay was 6 days and is double that reported in the US (post-operative stay 3 days in patients treated through the Medicare health insurance programme) [10]. Our analysis of LOS was confined to men treated in public hospitals only and those treated in public and private hospitals differed in terms of age, smoking status and deprivation category. Although none of these factors were significantly associated with prolonged LOS, it is possible that men treated privately might differ from those treated publicly in other ways that do impact LOS. For example, men treated privately might have fewer or less important co-morbidities or be fit for surgery. Therefore, our results confined to public patients could over-represent sicker or less fit patients or those with more complex disease, therefore driving up median LOS. It is impossible to assess whether this is true, however, since there are no data available on comorbidities, fitness for surgery, or case complexity in private patients.

In the US there has been a concerted effort to reduce post-surgery LOS following RP using collaborative care pathways [32‐34]. Further reductions have been driven by increased use of laparoscopic (LRP) and robot assisted RP (RARP) [10]. Unfortunately we did not have information on the type of RP conducted and so could not investigate the impact of these further. However, RARP was not available in public hospitals during the study period. It is not particularly usual for public hospitals in Ireland to document local guidelines for practice, so we would expect that most hospitals would not have established standard post-operative care pathways. Nor has there been any specific national initiatives to reduce LOS during the study period. This, together with differences in casemix, type of procedure performed and public and private healthcare cultures, probably explains the higher LOS in Ireland than in the US.

There was a modest downward trend in median LOS over the study period, similar to patterns reported elsewhere. In England, for example, Hanchanale et al. [9] reported a decrease in median LOS from 8 days (IQR, 7–10) in 1998 to 6 days (IQR, 5–8) in 2004. The authors reported a modest increase in the use of LRP but this accounted for only 4% of RP cases overall and was considered unlikely to fully explain the downward trend.

A range of national clinical initiatives to improve patient care post-surgery and reduce LOS are only currently being developed in Ireland [35] so cannot have impacted on our study results. However there may have been local initiatives of which we were not aware. Other possible – and plausible – explanations for the observed trend are greater surgical experience resulting from the increased volume of RPs carried out, together with improvements in anaesthetic techniques leading to faster post-operative recovery times [36].

In the adjusted analysis men who were not married, and who had co-morbidities or later stage disease were significantly more likely to have a prolonged hospital stay (>9 days). We found similar associations in patients with colorectal cancer who underwent resection [37]. The observation that married patients have shorter LOS may reflect a lack of social support among unmarried patients [38, 39]. Urologists may be reluctant to discharge men who have no identifiable family caregiver. A shortage of step-down beds could exacerbate this problem as patients may occupy hospital beds for longer than required [40]. It might be expected a priori that those with co-morbidities would have longer LOS; these patients may need more investigation and work-up before surgery or a longer recovery time post-surgery. Similarly patients with more advanced disease are likely to need more investigation and work-up before surgery, more extensive surgery and more investigation post-surgery to classify surgical margins and to be assessed for further treatment.

The observed associations between LOS and provider volume have been previously reported [32]. What seems to be uncertain is whether surgeon or hospital volume, or both, is important. Like us, in an analysis of NHS data in England, Hanchanale et al. [9] found both hospital and surgeon volume were inversely associated with LOS. However, Judge et al. [8], using the same source data, found that trust volume alone was important. The need for minimum volume thresholds for urological cancer surgery has been recognised by surgeons [41] and in England the National Institute for Clinical Excellence (NICE) recommends that RP should be provided by specialised teams typically carrying out 50 or more operations per annum [42]. In Ireland only two public hospitals conducted 50 or more RPs per year. Evidence also suggests that experience in a particular mode of RP (i.e. ORP, LRP, RARP) is important for improved patient outcomes [41, 43, 44].

Although we found that both hospital and surgeon volume were statistically independent predictors of LOS, in practice these are likely to be related; higher volume centres are more likely to have more experienced surgeons who in turn may drive quality improvements or improvements in perioperative care. However, higher-volume centres may also deal with more complex cases which cannot be adjusted for fully, using simple co-morbidity counts [45]. Unfortunately we do not have information on other predictors of case complexity (e.g. frailty), and so we may have underestimated the strength of association between higher volume and lower LOS.

There was a high rate of overnight elective readmissions following RP and, in further analysis, wide variations between public hospitals were evident. Some private hospitals in Ireland have a stated policy of overnight readmission for catheter removal and it is reasonable to assume that public hospitals with high overnight elective readmission rates have a similar policy. The cost-benefit implications for the health services nationally (and for individual hospitals) of overnight readmission versus elective day-patient care for catheter removal are unclear and warrant further investigation.

Although higher-volume hospitals had a higher overnight elective readmission rate than lower-volume hospitals, the risk of prolonged LOS remained lower in higher-volume hospitals when aggregated LOS (i.e. the sum of LOS for the RP itself and any readmissions within 28 days), rather than the surgery episode LOS, was considered. Although the overall numbers were low, and so need to be interpreted with caution, the data suggested that higher-volume hospitals had more emergency readmissions than lower-volume hospitals; this is probably a reflection of more complex cases being treated at higher-volume hospitals.

In comparison with the UK and US the total number of RPs undertaken in Ireland is small. In addition, the pool of hospitals and surgeons providing the service is small and this is further complicated by the public/private divide. The fact that 50% of RPs in public hospitals were carried out at two hospitals (and by eight urologists) demonstrates the diffusion of cancer services in Ireland. The National Cancer Control Programme (NCCP) has embarked on a programme of centralization of public cancer services. Greater centralization of prostate cancer services could facilitate the development of a pool of experienced surgeons potentially leading to a more robust service less sensitive to loss of personnel and better equipped to implement technological advances in RP surgery such as LRP and RARP. It would facilitate the implementation of clinical initiatives already shown elsewhere to improve patient and hospital-related outcomes (e.g. standardized critical care pathways) and – as suggested by this study – lead to shorter LOS. It would also enable investigation of other aspects of service provision that could both improve quality-of-care and might reduce costs (e.g. more efficient discharge policies and standardised follow-up care). However, centralisation would be likely to mean that patients would need to travel further to access care causing increased inconvenience and cost and, perhaps, impacting on their satisfaction with care. The relationship between these factors is complex, and a full analysis of cost and benefits is required.