8
RETRACTION Journal of Surgical Oncology Factors Affecting Hospital Readmission Rates For Breast Cancer Patients in Western Australia MICHAEL A. MARTIN, PhD,* RAMONA MEYRICKE, BSc, TERRY O’NEILL, PhD, AND STEVEN ROBERTS, PhD School of Finance and Applied Statistics, Australian National University, Canberra, Australia Background and Objectives: Unplanned hospital readmissions following surgical treatment for breast cancer are an indicator of morbidity. We explore the relationship between the risk of unplanned hospital readmissions and various factors, including tumor size and histology, lymph node involvement, the type of surgical treatment, mastectomy or breast-conserving surgery (BCS), and patient demographics. Methods: Linked Western Australian cancer mortality and hospital morbidity data were used in the assessment of readmissions following initial surgical treatment for breast cancer. Planned admissions for adjuvant treatment such as chemotherapy or radiotherapy were deleted. Survival models for multiple events per subject were applied to analyze the data. Results: Hazard of unplanned readmission rises by a factor of 1.005 for each mm in tumor size, is reduced by about 40% for metropolitan residents over rural-based patients, and by 4% for patients whose initial surgical treatment was BCS over mastectomy patients. Area of residence interacts with other factors, including patient age, lymph node involvement, and tumor histology. Conclusions: While use of BCS appears associated with lower long-term rates of unplanned hospital readmissions than those following mastectomy, the roles of other factors remain important. Patients living in metropolitan areas have significantly lower rates of readmission than rural/remote counterparts. J. Surg. Oncol. ß 2008 Wiley-Liss, Inc. KEY WORDS: breast-conserving surgery; breast cancer; conditional survival model; hospital readmission; mastectomy; post-operative morbidity; recurrent event analysis INTRODUCTION Breast cancer is a disease that affects about one tenth of Australian women. Because of its devastating impact on the community, much research has been conducted on multiple aspects of the condition, including possible causative factors, methods of treatment and patient care, and measures such as breast screening. For example, a number of clinical trials have concluded that breast cancer patients treated with breast-conserving surgery (BCS) versus mastectomy have similar long-term survi- val rates [1–3]. This evidence is supported by a recent pooled analysis of six major clinical trials comparing mastectomy and BCS, which concluded that the two treatments have comparable effects on mortality, even after long-term follow up [4]. Other population-based studies have also shown that patients receiving mastect- omy and BCS have similar survival rates [5,6]. Apart from survival rates associated with the breast cancer treatment options of BCS and mastectomy, another question of great importance for breast cancer patients is morbidity following surgical treatment for breast cancer. For example, even though BCS and mastectomy have been shown to have similar long-term survival rates it may be the case that one form of treatment affords a lower morbidity experience (e.g., fewer complications or better self-image), which if known to breast cancer patients facing a treatment choice may play an integral part in their choice of treatment. *Correspondence to: Michael A. Martin, PhD, School of Finance and Applied Statistics, Australian National University, Canberra ACT 0200, Australia. Fax: þ61-2-6125-0087 E-mail: [email protected] Received 24 May 2006; Accepted 9 November 2006 DOI 10.1002/jso.20742 Published online in Wiley InterScience (www.interscience.wiley.com). ß 2008 Wiley-Liss, Inc.

Factors affecting hospital readmission rates for breast cancer patients in Western Australia

Embed Size (px)

Citation preview

Page 1: Factors affecting hospital readmission rates for breast cancer patients in Western Australia

RETRACTIO

N

Journal of Surgical Oncology

Factors Affecting Hospital Readmission Rates ForBreast Cancer Patients in Western Australia

MICHAEL A. MARTIN, PhD,* RAMONA MEYRICKE, BSc, TERRY O’NEILL, PhD, AND STEVEN ROBERTS, PhD

School of Finance and Applied Statistics, Australian National University, Canberra, Australia

Background and Objectives: Unplanned hospital readmissions following surgicaltreatment for breast cancer are an indicator of morbidity. We explore the relationshipbetween the risk of unplanned hospital readmissions and various factors, includingtumor size and histology, lymph node involvement, the type of surgical treatment,mastectomy or breast-conserving surgery (BCS), and patient demographics.Methods: Linked Western Australian cancer mortality and hospital morbidity datawere used in the assessment of readmissions following initial surgical treatment forbreast cancer. Planned admissions for adjuvant treatment such as chemotherapy orradiotherapy were deleted. Survival models for multiple events per subject wereapplied to analyze the data.Results: Hazard of unplanned readmission rises by a factor of 1.005 for each mm intumor size, is reduced by about 40% for metropolitan residents over rural-basedpatients, and by 4% for patients whose initial surgical treatment was BCS overmastectomy patients. Area of residence interacts with other factors, including patientage, lymph node involvement, and tumor histology.Conclusions: While use of BCS appears associated with lower long-term rates ofunplanned hospital readmissions than those following mastectomy, the roles of otherfactors remain important. Patients living in metropolitan areas have significantly lowerrates of readmission than rural/remote counterparts.J. Surg. Oncol. � 2008 Wiley-Liss, Inc.

KEY WORDS: breast-conserving surgery; breast cancer; conditional survivalmodel; hospital readmission; mastectomy; post-operative morbidity;

recurrent event analysis

INTRODUCTION

Breast cancer is a disease that affects about one tenthof Australian women. Because of its devastating impacton the community, much research has been conducted onmultiple aspects of the condition, including possiblecausative factors, methods of treatment and patient care,and measures such as breast screening. For example, anumber of clinical trials have concluded that breastcancer patients treated with breast-conserving surgery(BCS) versus mastectomy have similar long-term survi-val rates [1–3]. This evidence is supported by a recentpooled analysis of six major clinical trials comparingmastectomy and BCS, which concluded that the twotreatments have comparable effects on mortality, evenafter long-term follow up [4]. Other population-basedstudies have also shown that patients receiving mastect-omy and BCS have similar survival rates [5,6].

Apart from survival rates associated with the breastcancer treatment options of BCS and mastectomy,another question of great importance for breast cancerpatients is morbidity following surgical treatment forbreast cancer. For example, even though BCS andmastectomy have been shown to have similar long-termsurvival rates it may be the case that one form oftreatment affords a lower morbidity experience (e.g.,fewer complications or better self-image), which ifknown to breast cancer patients facing a treatment choicemay play an integral part in their choice of treatment.

*Correspondence to: Michael A. Martin, PhD, School of Finance andApplied Statistics, Australian National University, Canberra ACT 0200,Australia. Fax: þ61-2-6125-0087 E-mail: [email protected]

Received 24 May 2006; Accepted 9 November 2006

DOI 10.1002/jso.20742

Published online in Wiley InterScience (www.interscience.wiley.com).

� 2008 Wiley-Liss, Inc.

Page 2: Factors affecting hospital readmission rates for breast cancer patients in Western Australia

RETRACTIO

N

Along these lines, a recent study showed significantlygreater levels of psychological morbidity in women whohad BCS compared with those who had mastectomy [7].Several other studies have also investigated the relativepsychosocial morbidity of mastectomy versus BCS, witheither equivocal [8], mixed [9–11] or even opposite [12]results with respect to a variety of psychosocial measuressuch as body image, anxiety about recurrence, and abilityto conduct daily activities. However, to the best of ourknowledge the issue of physical morbidity for breastcancer patients treated with BCS compared with thosetreated with mastectomy has received little attention inthe literature. This paper will attempt to address thisshortcoming in the literature as well as to investigatewhich other factors impact physical morbidity followingsurgical treatment for breast cancer.The aim of this paper is to study the factors associated

with unplanned readmissions to hospital for WesternAustralian female breast cancer patients diagnosed after1/1/1995. The period under study ended 3/31/2002.Planned readmissions are usually for check-ups, oradditional adjuvant therapy, following a surgical treat-ment for breast cancer. Unplanned readmissions, on theother hand, reflect a need for medical treatment. They arean important indicator of patient morbidity. For example,a comparison of unplanned hospital admission ratesbetween breast cancer patients treated with BCS andmastectomy, allowing for other disease and patientcharacteristics, may provide valuable information onwhich surgical treatment is associated with fewerphysical complications and enable cancer patients tomake more informed decisions about their choice oftreatment. Further, knowledge of which demographicvariables appear associated with increased re-admissionrisk may assist in identifying patients at increased risk ofpost-treatment physical morbidity.

MATERIALS AND METHODS

Data for the study were sourced from linked admin-istrative data obtained from the Western Australian (WA)Department of Health Record Linkage unit. The datasetwas extracted from the WA Linked Database, a dynamiclinkage system linking three core data sources: the WACancer Registry (WACR), the WA Hospital MorbidityDatabase (WAHMD) and the WA Death Register. TheWAHMD contains comprehensive patient demographic,diagnosis, and procedure information for each hospitaladmission occurring in any WA hospital. The datasetused consists of the linked hospital, death and WACRrecords containing the diagnosis, subsequent admissionsto hospital, and death (where applicable) of around 3,000women diagnosed with cancer in WA between 1 January,1995 and 31 December, 1999. Data prior to 1995 were

omitted because of the small number of observations withrecorded tumor size measurement before this time. TheWA Linked Database is unique in the Australian contextand is an extraordinarily rich and comprehensive resource[13].In this analysis, surgical treatment was defined as the

first procedure carried out following a diagnosis of breastcancer. The date of diagnosis is defined for this study asthe time at which the subject enters the study as recordedon the WACR as none of the databases linked by theWestern Australian unit specifies diagnosis date. The dateof first surgical treatment for breast cancer linked to eachdiagnosis was recorded on the WAHMD, and is used inthis study as the start time for each patient in this study.The explanatory variables included in the analysis werethe first surgical treatment received by the subject(mastectomy or BCS), tumor size (diameter, measuredin millimeters), nodal status (a categorical variable withcategories ‘‘no nodal involvement,’’ ‘‘1–3 nodes,’’ and‘‘4 or more nodes’’—see [14], [15]), tumor histology(a categorical variable classifying the tumor as ductal,lobular, or other), the age of the subject in years, thesubject’s area of residence (rural/remote or metropolitan),the subject’s country of birth (Australia/NZ or not),marital status, whether or not they were Aboriginal orTorres Strait Islander, and the method of payment(public, private, or no health insurance). Table I containssummary information for each of these variables.The extracted data constituted 2,703 patients. Of the2,703 patients in the study, 106 did not have informationon the nodal status variable described above. Of these106 patients, the majority (103) had large tumors(in excess of 50 mm), and so the results of our modelingare to be interpreted as restricted to tumors within 50 mmin diameter, a value of clinical significance as it reflectsgenerally accepted clinical guidelines relating to tumorsizes beyond which breast-conserving therapy is contra-indicated.A program was written in SAS that extracted the dates

of all unplanned hospital readmissions from theWAHMD. Planned readmissions fail to consistentlyreflect morbidity as it is common for patients with breastcancer to have multiple planned readmissions followinginitial surgery for adjuvant therapy such as chemotherapyor radiotherapy. Crude rates of readmission are, thus, notuseful because such planned readmissions account fora large proportion of readmissions. Therefore, priorto analysis we removed planned admissions from theextracted dataset. We identified planned admissions byscanning the separation record on the WAHMD forprocedure codes (from ICD-9 and ICD-10) that representsurgical after-care procedures such as chemotherapy,radiotherapy, occupational or physio-therapy, or proce-dure codes that represent a check-up or follow-up

Journal of Surgical Oncology

2 Martin et al.

Page 3: Factors affecting hospital readmission rates for breast cancer patients in Western Australia

RETRACTIO

Nprocedure. These codes were identified using both ICD-9and ICD-10 codings. Any separation that included oneor more of these codes was classified as a plannedreadmission and was removed from the dataset. All otherreadmissions were classified as unplanned. As a result,under the definition of unplanned admissions used here,while the bulk of readmissions were likely to haveresulted from complications related to breast cancertreatment, unrelated readmissions (e.g., for trauma andother illnesses) may have also been captured. While itmay have been preferable to restrict attention only toreadmissions explicitly related to breast cancer treatment,the lack of sufficiently specific ICD codings for a numberof admissions made such a choice difficult to implementreliably. Although we have simply used unplannedhospital readmissions as a proxy for physical morbidity,there are potentially better, more specific morbidityindicators that might be considered. Such examplesinclude breast and axillary wound infection after BCS orafter mastectomy, seroma formation requiring aspiration,restricted shoulder range of motion requiring physicaltherapy or rehabilitation, breast, or arm edema (short- andlong-term) requiring treatment, either by physical therapyor sleeve, margin re-excision after BCS, conversion fromBCS to mastectomy, and complications of reconstructionafter mastectomy. We elected to persist with the broaderunplanned readmission indicator as sufficiently specificICD codings were unavailable for a significant proportionof patients in the study. We note that hospital readmissionis more common in the Australian setting than in the

United States where certain procedures are morecommonly handled on an outpatient basis.

Our analysis of unplanned hospital readmissionsinvolves the application of survival analysis methods todata with multiple or recurrent events per subject. Thisapplication uses an extension of the Cox proportionalhazards model to allow for intrasubject correlationinherent in the data [16]. The correlation derives fromthe fact that the probability of a given subject having ahospital readmission, as well as the timing of thisreadmission, is dependent on the number and timing ofthe previous hospital readmissions for the subject inquestion. The model used in this paper was proposed byPrentice, William, and Peterson [17] and will be referredto as the PWP model. The PWP model was selectedbecause, in addition to handling the intrasubject correla-tion inherent in the data, it successfully handles theordered nature of the hospital readmission data; that is,the model allows for the fact that a subject cannot be atrisk for their kth hospital readmission until they haveexperienced k � 1 previous readmissions. The PWPmodel is described in detail in Therneau and Grambsch[16], where it is also referred to as the conditional model.The model is particularly suited to this type of data.

The difference between the Cox proportional hazardsmodel and the PWP model can be seen by consideringhow the two models model the hazard function. The Coxmodel specifies the hazard for subject i as:

l0ðtÞ expðXiðtÞbÞ

Journal of Surgical Oncology

TABLE I. Summary Statistics for Subject Characteristics, N¼ 2,703

Surgery Aboriginality Marital status

Mastectomy 970 (35.9%) No 2,680 (99.1%) Married/de-facto 1,867 (69.1%)

Breast-conserving surgery (BCS) 1,733 (64.1%) Yes 23 (0.9%) Otherwise 836 (30.9%)

Method of payment Country of birth Area of residence

Public, eligible for Medicare 1,340 (49.6%) Australia 1,651 (61.1%) Rural /remote 585 (21.6%)

Private, not insured 89 (3.3%) Other 1,052 (38.9) Metropolitan 2218 (78.4%)

Private, insured 1234 (45.7%)

Ineligible for assistance 40 (1.5%)

Lymph node involvement Tumor size (diameter in mm) Subject age

Node negative 1,565 (57.9%) Minimum 1 Minimum 23

Node positive (1–3 nodes) 700 (25.9%) 1st Quartile 12 1st Quartile 51

Node positive (4þ nodes) 332 (12.3%) Median 18 Median 59

Missing 106 (3.9%) 3rd Quartile 25 3rd Quartile 69

Maximum 190 Maximum 98

Tumor histology TNM status Survival

Ductal 2,151 (79.6%) T1 1,701 (62.9%) Survived 2,478 (91.7%)

Lobular 331 (12.2%) T2 899 (33.3%) Died 225 (8.3%)

Other 221 (8.2%) T3 103 (3.8%)

For categorical variables, the percentages of subjects in each category are presented. For continuous variables, five-number summaries are

presented.

Hospital Readmission Post-Breast Cancer 3

Page 4: Factors affecting hospital readmission rates for breast cancer patients in Western Australia

RETRACTIO

N

where l0ðtÞ is the baseline hazard, XiðtÞ are the covariatevalues for subject i, and b is the vector of coefficients.This model only allows for one event per subject. ThePWP model models the hazard function of the jth eventfor subject i as:

YijðtÞ ¼ l0jðtÞ expðXiðtÞbjÞ

where YijðtÞ is zero until the (j�1)th event occurs andthen becomes one, l0jðtÞ is the baseline hazard of the jthevent, and bj is the vector of coefficients for the jth event.From these two hazard functions, it is evident that thePWP model, unlike the Cox model, allows for multipleevents by allowing the hazard function to differ withevent number. Event numbers are treated as strata by thefitting algorithm, with a different baseline hazard usedwithin each stratum.To implement the PWP model, each subject is

represented as a series of observations with timeintervals:ð0; T1�; ðT1;T2�; . . . ; ðTN ;minðDi;CiÞ�, whereTi is the time of the ith hospital readmission, Di isthe time of death of subject i in the event that subject idies, while Ci is the time at which subject i ceased to beunder observation in the event that subject i is censored.By including death as a competing event, the PWP modelis able to produce a global assessment of treatment effectwith respect to both hospital readmission and a patient’smortality risk. The PWP model was implemented usingthe statistical package S-Plus [18]. For our application,the majority of subjects experienced four or fewerunplanned hospital readmissions. Table II summarizesthe distribution of number of readmissions for patients inour study. The table also presents this information brokendown by two factors: the TNM status of the tumor and thetype of surgical treatment used. The dataset did notcontain explicit information on the tumor TNM status.However, using information on tumor size we classifiedeach tumor as T1 (maximum diameter 2 cm or less), T2(diameter over 2 cm but not more than 5 cm), or T3(diameter exceeding 5 cm). Our dataset did not specify iftumors had, for example, penetrated the breast wall, and

so we were unable to classify tumors as T4. Patients with10 or more readmissions are rare, and this phenomenonraises difficulties in fitting the PWPmodel since the smallnumber of observations per stratum causes convergenceproblems for the fitting algorithm, which estimates aseparate baseline hazard for each stratum. In order toaddress this issue, patients having 10 or more read-missions were amalgamated into a single stratum forwhich a single, common baseline hazard was assumed.This approach, termed an amalgamated PWP model,allowed all model parameters to be estimated stably.

RESULTS

An initial amalgamated PWP model including allcovariates and two-way interactions was fit to the data.The PWP model makes a proportional hazards assump-tion within each event stratum, and a diagnostic check ofthis assumption using the Schoenfeld residuals from theinitial model fit revealed that the proportional hazardsassumption was reasonable with respect to all includedvariables except subject age. To account for this issue, aquadratic term in subject age was incorporated into themodel, and further diagnostic checks revealed theproportional hazards assumption within each stratumthen appeared reasonable. Following this model fit,several covariates and interactions were removed fromthe model as statistically non-significant, includingaboriginality, marital status, payment class and countryof birth and associated interactions, as well as otherinteraction terms whose P-value exceeded 0.05. Detailsof the final model fit are summarized in Table III.Standard errors of coefficients used in calculating z scoresand P-values were obtained using a robust jackknifeapproach outlined in Therneau and Grambsch [16]. Thechosen model contains both main effects and interactions,the combination of which paint a rich, detailed picture ofhow the relevant factors are associated with unplannedhospital readmissions. The broad interpretations arisingfrom the model described in Table III are summarized inTable IV.

Journal of Surgical Oncology

TABLE II. Distribution of Number of Unplanned Hospital Readmissions, N¼ 2,703

Number of

readmissions Overall

TNM status T1

(diameter� 2 cm)

TNM status T2

(2 cm< diameter� 5 cm)

TNM status T3

(diameter> 5 cm)

Mastectomy BCS Mastectomy BCS Mastectomy BCS

0 826 (30.6%) 116 (27.4%) 450 (35.2%) 131 (27.5%) 98 (23.2%) 28 (40%) 3 (9.1%)

1 695 (25.7%) 114 (27%) 323 (25.3%) 124 (26%) 116 (27.5%) 11 (15.7%) 7 (21.2%)

2 438 (16.2%) 71 (16.8%) 193 (15.1%) 86 (18%) 71 (16.8%) 11 (15.7%) 6 (18.2%)

3 280 (10.4%) 40 (9.5%) 136 (10.6%) 38 (8%) 55 (13%) 6 (8.6%) 5 (15.2%)

4 155 (5.7%) 17 (4%) 69 (5.4%) 35 (7.3%) 26 (6.2%) 4 (5.7%) 4 (12.1%)

5þ 309 (11.4%) 65 (15.4%) 107 (8.4%) 63 (13.2%) 56 (13%) 10 (14.2%) 8 (24.2%)

The number of re-admissions was further broken down by TNM category and first surgical treatment for breast cancer (mastectomy or BCS).

4 Martin et al.

Page 5: Factors affecting hospital readmission rates for breast cancer patients in Western Australia

RETRACTIO

NThe model fit reveals that patients more likely to

experience higher recurrence of unplanned hospitalreadmissions include patients with larger tumors, andthose whose primary treatment for breast cancer was BCSrather than mastectomy. The model includes multiple

interaction terms involving area of residence, althoughgenerally metropolitan-based patients appear to experi-ence about 60% of the baseline hazard of readmission oftheir rural/remote-based counterparts. Significant inter-actions exist between area of residence and lymph node

Journal of Surgical Oncology

TABLE III. Summary of Results From the Selected PWP Model Fit to the Hospital Readmission Data

Covariate Coef SE (coef) Exp (coef) P-value

Tumor size (effect per mm) 0.004996 0.0014 1.005 0.0027*

Tumor histology—ductal (baseline) — — — —

Tumor histology—lobular �0.04409 0.0264 0.957 0.097

Tumor histology—other 0.039805 0.0197 1.041 0.061

Area of residence—rural/remote (baseline) — — — —

Area of residence—metropolitan �0.507424 0.2356 0.602 0.041*

Lymph node negative (baseline) — — — —

Lymph node positive (1–3 nodes) �0.00776 0.0192 0.992 0.7

Lymph node positive (4þ nodes) 0.055911 0.0144 1.058 0.0012*

Subject age �0.046742 0.0082 0.954 <0.0001*

(Subject age)2 0.000455 0.00007 1 <0.0001*

Surgical treatment—mastectomy (baseline) — — — —

Surgical treatment—BCS �0.040223 0.0152 0.961 0.024*

Tumor histology—area interaction (ductal/rural; baseline) — — — —

Tumor histology—area metro interaction (lobular) 0.034734 0.0263 1.035 0.19

Tumor histology—area metro interaction (other) �0.082326 0.0197 0.921 <0.0001*

Lymph node negative—area metro interaction (baseline) — — — —

Lymph node positive—area metro interaction (1–3 nodes) 0.049568 0.019 1.051 0.013*

Lymph node positive—area metro interaction (4þ nodes) �0.007931 0.0139 0.992 0.64

Subject age/age2—area rural interaction (baseline) — — — —

Subject age—area metro interaction 0.01782 0.0082 1.018 0.042*

(Subject age)2—area metro interaction �0.000175 0.00007 1 0.018*

Tumor size is measured as diameter of tumor in millimeters, area of residence was coded 0 for rural/remote and 1 for metropolitan, first surgical

treatment was coded as 0 for mastectomy and 1 for BCS, lymph node status was a factor with three levels (0 nodes, 1–3 nodes, and 4þ nodes), and

tumor histology was a factor with three levels (ductal, lobular, and other). Terms significant at the 5% level are indicated with an asterisk (*).

TABLE IV. Interpretation of Final PWP Model Fit (Effects of Significant Predictors as a Proportion of the Baseline Hazard ofReadmission)

Tumor size Hazard of readmission increases with tumor size, by a factor of 1.005 for each mm increase in size

(effect statistically significant); no significant interactions exist involving tumor size.

Surgical treatment

(mastectomy or BCS)

Hazard of readmission is reduced by 4% for BCS patients compared with mastectomy patients

(effect statistically significant); no significant interactions exist involving this variable.

Area of residence Average reduction in hazard is substantial—roughly 40%, with metro patients having about 60% of the

hazard of rural patients (averaging over factors with which area of residence interacts); however,

area interacts with a number of other variables (see below)

Tumor histology ductal/

lobular/other

Patients with ductal tumors who live in rural areas are the baseline; relative hazards are presented in the

following table (statistically non-significant results are in italics)

Interaction with area of residence Ductal Lobular Other

Area rural Baseline 0.957 1.041

Area metropolitan 0.602 0.596 0.577

Nodal involvement

(none/1–3/4þ)

Patients with no nodal involvement residing in rural areas are the baseline; relative hazards are presented in

the following table (statistically non-significant results are in italics)

Interaction with area of residence No nodal involvement 1–3 Nodes 4þ Nodes

Rural Baseline 0.992 1.058

Metropolitan 0.602 0.628 0.632

Subject age Generally, the hazard of readmission falls slightly with age until about age 51 whereupon it rises, for both

rural and metropolitan patients, although for metropolitan patients, the baseline is lower and the rise

post-age 51 is shallower; for rural patients, the hazard is proportional to exp (�0.0467 Ageþ 0.000455

Age2, while for metropolitan patients, the hazard is proportional to 0.602 exp (�0.0289 Ageþ 0.00028

Age2; Figure 1 displays the comparative hazards (rural is solid line, metro is dotted)

Hospital Readmission Post-Breast Cancer 5

Page 6: Factors affecting hospital readmission rates for breast cancer patients in Western Australia

RETRACTIO

N

involvement, area and tumor histology, and area andsubject age. The effects of these interactions in terms ofthe proportion of baseline hazard are presented in detailin Table IV.For example, considering the interaction between area

of residence and lymph node involvement, the extent towhich the number of involved lymph nodes affects thehazard of readmission differs between rural/remote-basedpatients and metropolitan-based patients. For rural/remote-based patients, the hazard does not changesignificantly between patients with no lymph nodeinvolvement and patients with 1–3 affected nodes, butrises by about 6% for patients with 4 or more affectednodes. In contrast, for metropolitan-based patients, thereis a statistically significant increase in the risk ofreadmission for patients with 1–3 nodes over patientswith no nodal involvement. The interaction between areaof residence and age is particularly interesting, with themodel suggesting a declining hazard between ages 20 and50 for both rural and metropolitan patients, after whichthe hazard rises for both groups, but with significantlyslower changes in the hazard for metropolitan patients.Also of interest is the fact that the predicted hazard islowest for both groups at roughly the same age (51),despite no such restriction being imposed by the model.Figure 1 displays the predicted relative hazard for ruraland metropolitan patients by age.

DISCUSSION

Several previous studies have considered morbidityassociated with breast cancer treatments in severalcountries. Both Taylor [18] and Schrenk et al. [19]consider morbidity associated with axillary lymph nodedissection as part of breast cancer treatment, andconclude that the procedure has significant, long-termmorbidity consequences for patients. More generally,Hynes et al. [20] showed that morbidity and readmission

rates following treatment for breast cancer did notdiffer significantly between Veterans Affairs hospitalsand private sector hospitals in the United States. In ourstudy, we found no significant effect on readmission ratesarising from patients choice of private health insuranceversus public health insurance (called Medicare inAustralia). A Singapore study explored a protocolreferred to as ‘‘The Mastectomy Clinical Pathway’’ andfound significantly lower hospital readmission rates forpatients taking part in that protocol than for other patients[21]. Nevertheless, these studies did not focus on theissue of how a patient’s choice of mastectomy or BCSimpacted long-term hospital readmission rates, adjustingfor other factors such as tumor characteristics and cancerstage as well as patient demographics.Some of the results of our modeling are reasonably

easily anticipated. For instance, patients with lymph nodeinvolvement are often at a more advanced stage of thedisease, and are hence more likely to suffer long-termmorbidity issues than patients at earlier stages. Similarly,it might be expected that patients with larger tumorswould also experience higher morbidity than patientswith smaller tumors. Note, however, that tumor size is animportant factor leading to the choice of first surgicaltreatment. Martin et al. [22] found that tumor size was aprimary determinant of type of surgical treatment using aclassification tree approach. They noted that higher ratesof mastectomy were found among patients with largertumors. Current clinical practice guidelines vary, some-times significantly, by country, but the National Healthand Medical Research Council (NHMRC) [23] inAustralia, in its Clinical Practice Guidelines for theManagement of Early Breast Cancer indicate [18, page52] that although ‘‘there is no absolute limit to the size ofa tumor which can be locally excised without incurring ahigh risk of recurrence; 3–4 cm is often regarded as apractical limit,’’ citing Calais et al. [24]. Certainly,patients with smaller tumors (less than 20 mm) indiameter generally experienced low unplanned read-mission rates in the context of our data set, and we alsoobserved relatively low unplanned readmission rates forpatients with tumors up to 30 mm in diameter. In theUSA, neither the most commonly quoted NationalComprehensive Cancer Network (NCCN) guidelines forbreast cancer care [25], nor the Institute for ClinicalSystems Improvement (ICSI) health care guidelinesfor breast cancer treatment [26] stipulate that there isan absolute size above which BCS should not be used,but both sets of guidelines indicate a relative contra-indication to breast-conserving therapy requiring radia-tion therapy for tumors beyond 5 cm (Category 2B) indiameter.The significance of area of residence on unplanned

readmission experience is interesting and has obvious

Journal of Surgical Oncology

Fig. 1. Relative hazard by subject age for rural- versus metropolitan-based patients.

6 Martin et al.

Page 7: Factors affecting hospital readmission rates for breast cancer patients in Western Australia

RETRACTIO

N

public health implications. Also, the fact that area ofresidence significantly interacts with other variables inour study is of interest as it indicates that the effect of areaof residence is not uniform with respect to other variablessuch as tumor histology and nodal status. Subjectsresiding in rural or remote areas were predicted to havesignificantly higher risk of unplanned readmission thantheir metropolitan counterparts with a baseline reductionin the hazard of around 40% associated with metropolitanresidence even allowing for the significant interactionterms. This phenomenon perhaps reflects that suchpatients have fewer initial treatment options available tothem. An alternative explanation for the phenomenon isthat the complications of metropolitan residents mightoften be handled on an outpatient basis instead of by re-admitting them as in-patients, whereas patients in rural orremote locations may have geographic constraints whichmake such out-patient treatment infeasible. The predictedeffect of age on morbidity, for both metropolitan-basedand rural/remote-based patients, with hazard initiallydeclining and then rising after age 51 is also of interest.This effect may be related to the increased likelihood ofolder subjects choosing mastectomy as their surgicaltreatment, but may also be related to generally increasedmorbidity among older subjects.

It is also of interest to examine how the TNM stage forthe cancers impacts unplanned readmissions, in particularbetween the two surgical treatment groups. Table IIprovides details of number of unplanned readmissionsboth overall and broken down by TNM stage and type ofsurgical treatment. Within our data set, about 30% ofpatients overall have no unplanned readmissions follow-ing surgery, about 56% have one or fewer unplannedreadmissions, and nearly 75% have two or fewerreadmissions. Table II also shows that the number ofreadmissions for a patient depends critically on both theTNM stage and the type of surgical treatment. Forexample, patients with tumors classified as T1 who haveBCS as their primary treatment have significantly lowerreadmission rates than those having mastectomy, whilefor patients with tumors classified as T3, BCS patientshave a very poor readmission experience compared withmastectomy patients. For patients with tumors classifiedas T2, the readmission experiences of mastectomy andBCS patients are broadly similar. While these resultsseem easily anticipated, Table II quantifies the change inreadmission experience with respect to tumor stage andtype of treatment. Of course, treatment of T3 tumors withBCS is relatively rare, but seems to result in a fairly bleakreadmission experience.

One outcome of the study—that subjects having BCSas their first surgical treatment for breast cancer havestatistically significantly lower unplanned hospital read-mission rates than patients who have mastectomy, even

adjusting for factors such as tumor size, lymph nodestatus, tumor histology and patient demographics—suggests that the choice of BCS in the treatment ofbreast cancer offers patients the prospect of improvedmorbidity outcomes over the experience of patientswhose primary treatment is mastectomy. Of course, suchtreatment choices must be made in the context of manyother factors than post-treatment morbidity, and thisstudy does not address these issues. Moreover, the use ofunplanned hospital readmission as a proxy for morbidityignores other types of morbidity that may not be treatedvia hospital admission such as psychological impacts oftreatment, and so on.

It should be acknowledged that many of the covariatesused in our study are inter-related, and so may be thoughtof as ‘‘jointly’’ related to the risk of readmission ratherthan as acting in isolation. For example, variables relatingthe stage of the disease at presentation are likelydependent on the socio-economic status and level ofeducation of patients, and so the extent to which tumorsize/stage impact readmission risk is confounded with theeffects of other demographic variables is difficult toassess in practice.

Limitations of the present study must also be acknowl-edged. First, our study is based on data collected overonly a relatively short timeframe as data on all thecovariates included in our study were unavailable over alonger period. Also, while the Western Australiandatabase is unique in the Australian context in linkingcritical data from several sources, it still lacks informa-tion on certain variables (e.g., the prevalence of standardre-admission guidelines) that could be usefully consid-ered when modeling hospital readmission data. The issueof whether the threshold for re-admission was the sameacross different institutions and based on some sort ofstandard re-admission guidelines is of practical impor-tance because busy centers may have higher thresholds ascompared to centers with lower workloads. Issues relatedto specific post-surgical complications are also relevant tothe question of whether re-admission is necessary. Forexample, patients with wound infections after modifiedradical mastectomy are usually admitted as such a woundis in continuity with the axilla and the risk of severecomplications is very high. This is in contrast to patientswith breast conservation where the lumpectomy cavity isusually segregated well from the axillary cavity, and somanagement of such wounds can usually be safelyhandled without re-admission. Other areas where furtherresearch are warranted are the question of shorter-termimpacts of surgical treatment, and whether the increasedreadmission rates associated with mastectomy are forreconstructive surgery, complications arising from themastectomy or other medical treatment required in thelong term.

Journal of Surgical Oncology

Hospital Readmission Post-Breast Cancer 7

Page 8: Factors affecting hospital readmission rates for breast cancer patients in Western Australia

RETRACTIO

N

Despite these limitations, our analysis describes howvarious factors such as disease stage, type of surgicaltreatment, and patient demographics affect unplannedhospital readmissions—an indicator of physical morbid-ity—following surgical treatment for breast cancerpatients.

REFERENCES

1. Veronesi U, Saccozzi R, Del Vecchio M, et al.: Comparing radicalmastectomy with quadrantectomy, axillary dissection, and radio-therapy in patients with small cancers of the breast. N Engl J Med1985;312:665–673.

2. Fisher B, Redmond C, Poisson R, et al.: Eight-year results of arandomised clinical trial comparing total mastectomy andlumpectomy with or without irradiation in the treatment of breastcancer. N Engl J Med 1994;320:822–828.

3. Jacobson JA, Danforth DN, Cowan KH, et al.: Ten-year results ofa comparison of conservation with mastectomy in the treatment ofstage I and II breast cancer. N Engl J Med 1995;332:907–911.

4. Ismail J, Proschan MA: Randomised trials of breast-conservingtherapy versus mastectomy for primary breast cancer: A pooledanalysis of updated results. Am J Clin Oncol 2005;28:289–294.

5. Lee-Feldstein A, Anton-Culver H, Feldstein PJ: Treatmentdifferences and other prognostic factors related to breast cancersurvival. Delivery systems and medical outcomes. JAMA 1994;271:1163–1168.

6. Winchester DJ, Menck HR, Winchester DP: The National CancerDatabase report on the results of a large nonrandomisedcomparison of breast preservation and modified radical mastect-omy. Cancer 1997;80:162–167.

7. Cohen L, Hack TF, de Moor C, et al.: The effects of type ofsurgery and time on psychological adjustment in women afterbreast cancer treatment. Ann Surg Oncol 2000;7:424–434.

8. Sanger CK, Reznikoff M: A comparison of the psychologicaleffects of breast-saving procedures with the modified radicalmastectomy. Cancer 1981;10:2341–2346.

9. Nano MT, Gill PG, Kollias J, et al.: Psychological impact andcosmetic outcome of surgical breast cancer strategies. ANZ J Surg2005;75:940–947.

10. Noguchi M, Saito Y, Nishijima H, et al.: The psychological andcosmetic aspects of breast-conserving therapy compared withradical mastectomy. Surgery Today—The Japanese Journal ofSurgery 1993;23:598–602.

11. Yeo W, Kwan WH, Teo PML, et al.: Psychological impact ofbreast cancer surgeries in Chinese patients and their spouses.Psycho-Oncology 2004;13:132–139.

12. Al-Ghazal SK, Fallowfield L, Blamey RW: Comparison ofpsychological aspects and patient satisfaction following breastconserving surgery, simple mastectomy and breast reconstruction.Eur J Cancer 2000;36:1938–1943.

13. Giles GG: Medical record linkage in Australia: This is as good asit gets. ANZ J Surg 2005;75:259.

14. Foo CS, Su D, Chong CK, et al.: Breast cancer in young Asianwomen: Study on survival. ANZ J Surg 2005;75:566–572.

15. Jayasinghe UW, Taylor R, Boyages J: Is age at diagnosis anindependent prognostic factor for survival following breastcancer? ANZ J Surg. 2005;75:762–767.

16. Therneau TM, Grambsch PM:Modelling survival data: Extendingthe cox model. New York: Springer-Verlag, 2000.

17. Prentice RL, Williams BJ, Peterson AV: On the regressionanalysis of multivariate failure time data. Biometrika 1981;68:373–379.

18. Taylor KO.: Morbidity associate with axillary surgery for breastcancer. ANZ J Surg 2004;74:314–317.

19. Schrenk P, Rieger R, Shamiyeh A, et al.: Morbidity followingsentinel lymph node biopsy versus axillary lymph node dissec-tion for patients with breast carcinoma. Cancer 2000;83:608–614.

20. Hynes D, Weaver F, Morrow M, et al.: Breast cancer surgerytrends and outcomes: Results from a national Department ofVeterans Affairs study. J Am Coll Surg 2004;198:707–716.

21. Santoso U, Iau PTC, Lim J, et al.: The mastectomy clinicalpathway: What has it achieved? Ann Acad Med Singapore 2002;31:440–445.

22. Martin MA, Meyricke R, O’Neill T, et al.: Mastectomy or breastconserving surgery? Factors affecting type of surgical treatmentfor breast cancer—A classification tree approach. BMC Cancer.2006;6:98.

23. National Health and Medical Research Council (NHMRC,Australia): Clinical practice guidelines for the management ofearly breast cancer: Second edition. 2001.

24. Calais G, Berger C, Descamps P, et al.: Conservative treatmentfeasibility with induction chemotherapy, surgery, and radio-therapy for patients with breast carcinoma larger than 3 cm.Cancer 1994;74:1283–1288.

25. National Comprehensive Cancer Network (NCCN, USA): BreastCancer: Clinical Practice Guidelines in Oncology, v.2. 2006.http://www.nccn.org/professionals/physician_gls/default.asp

26. Institute for Clinical Systems Improvement (ICSI, USA): HealthCare Guideline: Breast Cancer Treatment. 2005. http://www.icsi.org/knowledge/detail.asp?catID¼29&itemID¼154

Factors associated with lower rates of unplanned hospital read-

missions include smaller tumor size, metropolitan area of

residence, and initial surgical treatment being breast-conserving

surgery. Other relevant factors include patient age, lymph node

involvement and tumor histology.

Journal of Surgical Oncology

8 Martin et al.