Cost and resource utilization in cervical cancer management: a real-world retrospective cost analysis

Objectives We set out to assess the health care resource utilization and cost of cervical cancer from the perspective of a single-payer health care system. Methods Retrospective observational data for women diagnosed with cervical cancer in British Columbia between 2004 and 2009 were analyzed to calculate patient-level resource utilization patterns from diagnosis to death or 5-year discharge. Domains of resource use within the scope of this cost analysis were chemotherapy, radiotherapy, and brachytherapy administered by the BC Cancer Agency; resource utilization related to hospitalization and outpatient visits as recorded by the B.C. Ministry of Health; medically required services billed under the B.C. Medical Services Plan; and prescriptions dispensed under British Columbia’s health insurance programs. Unit costs were applied to radiotherapy and brachytherapy, producing per-patient costs. Results The mean cost per case of treating cervical cancer in British Columbia was $19,153 (standard error: $3,484). Inpatient hospitalizations, at 35%, represented the largest proportion of the total cost (95% confidence interval: 32.9% to 36.9%). Costs were compared for subgroups of the total cohort. Conclusions As health care systems change the way they manage, screen for, and prevent cervical cancer, costeffectiveness evaluations of the overall approach will require up-to-date data for resource utilization and costs. We provide information suitable for such a purpose and also identify factors that influence costs.


INTRODUCTION
Cervical cancer and its precursors cause significant morbidity and mortality in Canada each year 1 .Steady rates of screening uptake by Canadian women resulted in incidence and mortality rate reductions from 1972 to 2006 2 .Nevertheless, in 2014, an estimated 1450 Canadian women were diagnosed with cervical cancer, and 380 women died from their disease 3 .
Screening and management techniques for cervical cancer are changing as new technologies become available.The recognition that oncogenic human papillomavirus (hpv) infection is a necessary factor for virtually all cervical cancers has led to new approaches to treatment, screening, and prevention, including population-based hpv vaccination programs for young women [4][5][6] .Screening techniques are also changing to reflect a growing knowledge base about the role that hpv plays in cancer development [7][8][9] .Introducing the new methods has been shown to lower the incidence of cervical cancer, which will have implications for population health and the economic burden associated with the disease 9,10 .In an era of evidence-based policy and constrained budgets, those economic implications are likely to be important to health care decision-makers.High-quality evidence about the resource utilization and costs associated with cervical cancer management can be used to develop policies for cancer screening and management 2 .
The currently available scientific literature about the cost of the clinical management of cervical cancer is limited in usefulness because of the age of the available data, the non-transferability of costs derived from a specific jurisdiction, and the incomplete reporting of management costs [11][12][13] .For example, recent European studies on the cost of cervical cancer provide a very limited scope of management, because they have collected data on either the first year after diagnosis or in-hospital costs only [14][15][16] .Cost analyses for the full management of cervical cancer in the U.S. health care system use data from private insurance claims [17][18][19] .Given the limitations associated with those methods (transferability of costs, representativeness of international estimates, and age of the data), health care decision-makers can benefit from up-to-date and comprehensive estimates of resource utilization and costs when considering new approaches to screening and treatment.
The present study uses retrospective patient-level data to estimate resource utilization and costs associated with the management of patients diagnosed with cervical cancer.

METHODS Ethics
Approval for this study was provided by the University of British Columbia Research Ethics Board (REB approval no.H11-01138).

Study Cohort
The retrospective cohort of women treated for cervical cancer was created based on electronic records from the Cancer Information System (cais) maintained by the BC Cancer Agency (bcca), the agency primarily responsible for cancer treatments provided in British Columbia and the sole provider of chemotherapy and radiation therapy in the province.All treatments, appointments, and other services provided under the auspices of the bcca are recorded in cais.
Treatment for women with cervical cancer is typically managed by an oncologist at a clinic within the bcca.After curative treatment [usually some combination of hysterectomy, radiotherapy (rt), and systemic therapy], women are routinely monitored for evidence of disease relapse (or progression in the case of incurable disease).After 5 years of disease-free survival, women are discharged from care at the bcca and return to their family physician.Women presenting with incurable disease are managed palliatively.
All women with a record of diagnosed cervical cancer treated at the bcca between 1 January 2004 and 31 December 2009 were eligible for inclusion in the retrospective cohort.Cervical cancer was defined as invasive cancer forming in the cervix, classified as invasive using the International Classification of Diseases for Oncology behaviour code 3. Using cais, demographic and clinical information (age, sex, date of death, number and type of appointments and tests received in bcca cancer centres) was collected.Records were excluded if the woman had received any cancer treatment outside of British Columbia (based on physician notes available from a manual review of electronic charts, conducted by IC and ZF) or any experimental treatment (that is, they had participated in a clinical trial), or if she had been treated for more than one type of cancer within 5 years of the cervical cancer diagnosis.The period of data collection commenced with diagnosis and concluded with either death or discharge from the bcca.Records for women who were neither discharged from the bcca nor deceased as of 31 March 2012 were censored at that date, to allow for at least 1 year of follow-up data for all members of the cohort.
We used electronic patient charts to abstract clinically relevant demographic data, including International Federation of Gynecology and Obstetrics stage at diagnosis, treatment procedure, and recurrence date.Date of recurrence was defined as an oncologist-confirmed return of cancer after treatment and remission, abstracted directly from each patient's electronic medical record 20 .
In British Columbia, cervical cancer screening is managed provincially, through the bcca Cervical Cancer Screening Program (ccsp).That program, the first of its kind, was established in 1955 and, since commencing, has lowered cervical cancer rates by 70% 21 .The program reminds clinicians when women are due for screening, tracks adherence to the screening recommendations, and monitors the performance of the system and the outcomes of screening activities.Cervical cancer incidence and mortality rates remain low in the province, illustrating the value of an organized screening program 22 .
To determine whether costs are different for women who participate in regular screening and women who do not, data from the ccsp were used to identify women who had attended at least 1 screening appointment between 6 months and 5 years before a cervical cancer diagnosis-the ccsp's definition of active screening participation 23 .Screening in British Columbia uses conventional Pap cytology.

Costing and Resource Utilization
Electronic records were used to abstract resource utilization and cost data for chemotherapy, hospitalization, rt, brachytherapy, medical services covered by provincial insurance, and prescription medication.A health care system perspective was chosen for the analysis.
Type of chemotherapy, date of administration, number of cycles, drug costs, and number of chemotherapy lines per patient were provided by the bcca Systemic Therapy Program database.Delivery costs reflecting nursing, pharmacy, and administration components, provided by the bcca Systemic Therapy Program, were multiplied by the number of cycles and combined with drug costs to produce a per-patient chemotherapy cost.Documentation of external-beam rt and brachytherapy-including admission date and number of fractions or number of insertions-was taken from bcca records.To estimate the cost of external-beam rt, a bcca-defined per-fraction delivery cost of $325 (an estimate derived from budgetary numbers and annual throughput from the bcca's Vancouver cancer centre) was multiplied by the total number of fractions delivered to each patient.A per-insertion brachytherapy delivery cost of $3705 was estimated based on the cost of administrative overhead and supplies, plus salary of staff at the booking, planning, pre-implant computed tomography, and operating room stages of the procedure (French J. Personal communication, 15 November 2013).The per-fraction and per-insertion cost were both calculated according to expert opinion from bcca staff and clinicians.
Costs of hospitalization in this analysis were defined as the value of all medical treatment costs and resources accumulated while in hospital, including outpatient visits.Admission date, Canadian Classification of Health Intervention procedure code, hospital location, resource intensity weight, and World Health Organization International Classification of Disease diagnosis codes were retrieved from the Discharge Abstract Database maintained by the Canadian Institute of Health Information and held by the B.C. Ministry of Health.Hospital visits with diagnosis codes indicating cervical cancer as either the primary reason for hospitalization or as a comorbid condition (C53.0-C53.9,C57.9) were included in the study.
Resource intensity weight represents the expected resource consumption of patients with similar hospital procedures ("case-mix groups"), adjusted for age 24 .To estimate hospitalization costs, the resource intensity weight for each visit was multiplied by the cost of treating the average patient in the corresponding hospital ("cost per weighted case").
Costs and descriptions of provincially insured prescription medications (from PharmaNet) and insured medical services through the B.C. Medical Services Plan (msp) were collected from the Ministry of Health.The msp covers all services defined as medically necessary by the province of British Columbia and is roughly equivalent to insurance plans in other provinces (for example, the Ontario Health Insurance Program, the Régie de l'assurance maladie du Québec, the Alberta Health Care Insurance Plan, and so on).PharmaNet is a province-wide system that records all prescriptions filled in British Columbia at community pharmacies, including amounts dispensed and costs.
Reasons for the prescription of various non-chemotherapy drugs (that is, "drug X was prescribed because of cervical cancer") were not included in the available datasets, and so resource utilization by type of prescription is not presented.Similarly, resource utilization for msp services (that is, number of appointments, tests, or devices) could not be estimated because of the large number of medically required insured services and the numerous potential purposes for each service.Nevertheless, prescription and msp records clearly not pertaining to cervical cancer (for example, antihistamines, eye examinations) were excluded from the analysis a .
Datasets from the bcca and Ministry of Health were linked by personal health number (a number unique to each insured resident of British Columbia).All msp, PharmaNet, and hospital records with an undetermined association with cervical cancer-that is, any billed service that was not clearly unrelated to cervical cancer b -were included.Because of incomplete reporting in the relevant datasets, costs related to home or community care were not included in the analysis.Because PharmaNet and msp data are not reportable in useful discrete units, resource utilization for those datasets is not presented.

Subgroup Analysis
Analyses of the cohort by several clinical subgroups were conducted: n By clinical stage n By histologic type (adenocarcinoma, squamous cell carcinoma, other histology-that is, histology that was neither adenocarcinoma nor squamous cell carcinoma, including adenosquamous carcinoma and carcinoma not otherwise specified) n By age in quartiles n By experience of disease recurrence during the observation period (yes or no) n By confirmed date of death during the observation period (yes or no) n By record of participation in the ccsp between 6 months and 5 years before the diagnosis of cervical cancer (that is, "active participation" in screening)

Statistical Analyses
Resources used before the diagnosis and after the date of discharge were beyond the scope of the present analysis.All cost figures are reported in 2014 Canadian dollars, adjusted for inflation according to the Consumer Price Index 25 .Many members of the cohort (64%) did not reach an event endpoint (death or 5-year discharge).To estimate the 5-year costs for censored cohort members ("right censoring"), the improved (partitioned) survival-weighted estimate of mean cost was calculated according to the method proposed by Bang and Tsiatis 26 .Briefly, Kaplan-Meier survival analysis was used to account for the presence of right censoring in the data, providing an estimate of mean survival and mean cost.A program written for the SAS statistical software application (SAS Institute, Cary, NC, U.S.A.) was used to perform the survival and other statistical analyses 27 .Costs were grouped into 3-month intervals for the partitioned analysis.The proportion of the total cost was calculated as the ratio between the survival-weighted estimate of a given cost component (for example, the cost of chemotherapy) and the survival-weighted estimate of total cost.
A Kruskal-Wallis H-test was used to test for significant differences in resource utilization within each subgroup.Analysis of variance was used to test for significant differences in the proportion of the total cost that each component represented across subgroups.a The MSP and pharmacy records were deemed unrelated if they matched any of these descriptions: tobacco or alcohol cessation, antihistamines, dermatology drugs, cardiovascular medications, respiratory treatments (for example, for asthma, bronchitis, emphysema), psychiatric medications, ulcer medications, medications to treat sexually transmitted infections, glaucoma medications, medications for bacterial infections, antifungal medications, or contraceptives.b Hospital visits were deemed unrelated if they matched any of these descriptions: related to pregnancy or childbirth, diseases of the digestive and genitourinary system, mental health issues, musculoskeletal system (including falls, cataracts, and fractures), or epilepsy.

Patient Characteristics
The final cohort included 563 women (642 eligible, less 8 enrolled in clinical trials and 71 with multiple cancers) with a mean age at diagnosis of 51.4 ± 15.5 years.Table i presents descriptive statistics for the cohort.The statistically "average" cohort member had a stage i squamous cell carcinoma and was alive at the end of the observation period, having been discharged to her family physician or still being in post-treatment follow-up.

Resource Utilization
Table ii presents mean resource utilization for the full cohort and the defined subgroups.
Although mean values are helpful in understanding resource use, resources are consumed in indivisible units (that is, a fraction of a hospital visit is impossible).It might therefore be more sensible to think of average resource utilization in terms of median rather than mean values.For the 463 patients with any record of admission to hospital, the median length of stay for an inpatient visit was 4 days [interquartile range (iqr): 2-8 days], with a median of 1 inpatient visit (iqr: 1-2 visits).Median outpatient visits were 1 (iqr: 1-2).The reason for hospitalization was not directly observable in this retrospective review of clinical records; however, analysis of the hospital procedure codes suggested that the most commonly-used codes concerned the cervix or areas of common cervical metastasis.
The 332 patients treated with rt received a median of 25 fractions (iqr: 25-30 fractions).The 229 patients treated with brachytherapy received a median of 2 brachytherapy insertions (iqr: 2-4 insertions).The 245 patients treated with chemotherapy received a median of 5 cycles [iqr: 5-5 cycles (patients are treated according to written protocols, meaning that no great variation in the number of cycles each person receives is expected)].

Costs
Table iii presents mean costs, calculated using the Bang-Tsiatis partitioned estimate.The average cost from diagnosis to death or discharge from the bcca was $19,153 (standard error: $3,484) per person.Costs for the various clinical subgroups are also presented.Inpatient hospital visits, services billed to the provincial health insurance plan, and rt (including brachytherapy) were the largest components of the total cost.The relative proportions varied between the subgroups, particularly those for stage and age at diagnosis (Table iv).
Our results showed that management costs were lower for women who had been active screening program participants (screened 6 months to 5 years before diagnosis) than for those who had not, although the difference was not statistically significant.We urge readers to take great care in the interpretation of that association.Based on the data from the ccsp, a conclusion cannot necessarily be drawn concerning whether a given cancer was diagnosed because of regular participation in the screening program or whether screening appointments were made for other reasons (investigation of symptoms, for instance).Women who did not fall into the "active screening program participant" category can be considered to fall into 3 distinct groups: women who received a screening test 6 months before diagnosis (n = 128), women who had a screening appointment more than 5 years before diagnosis (n = 38), and women with no ccsp screening record (n = 145).

Survival
Table iii presents mean survival by the Kaplan-Meier method.Mean time on study in the total cohort, from diagnosis to death or discharge from the cancer agency, was 64.5 months (standard error: 2.5 months).Women with stage i cancer, adenocarcinoma, and in the youngest age quartile had the highest mean survival; women presenting with stage iv cancers and those in the oldest age quartile experienced a significantly shorter survival than average (p < 0.05).Compared with women who were not active screening program participants, those who were active participants experienced, on average, an additional 14.8 months of survival.Women who were regular participants in screening were also more likely to have a cancer diagnosed at stage i (75.9% vs. 41.6%,chi-square test p < 0.0001).

DISCUSSION
Using an analysis of cost and resource utilization in invasive cervical cancer among women diagnosed at the BC Cancer Agency, our study estimated cancer management costs at a per-patient level.The results can be helpful in providing contemporary cost and resource inputs for investigations of screening and management in cervical cancer.
Before conducting the analysis, we hypothesized that participation in regular screening might affect the likelihood of extra-cervical spread and therefore the treatment cost of cervical cancer.Although we felt that use of the last screening date is a reasonable proxy for participation in regular screening, we cannot directly conclude from our analysis that the cost of managing cancers diagnosed as a result of a regular screening appointment are or are not different from the cost of cancers diagnosed in other ways.Further investigation is necessary to determine whether cancers detected through population screening are different, in terms of cost, from cancers detected after the onset of symptoms or by some other method.
Chemotherapy costs were incomplete for 6 patients in our cohort, and removing those patients from the cohort had a negligible effect on the estimated median and mean costs.This clarification is provided for the purpose of transparency; given the small number of records excluded for that reason, we believe that our results still adequately represent the total population of cervical cancer patients in British Columbia.

Comparison with the Literature
Recent European studies of the cost of cervical cancer management did not account for management costs from diagnosis to death or discharge, or for costs outside the hospital, making comparisons with our study difficult [14][15][16] .
McCrory et al., Helms and Melnikow, and Insinga et al. used U.S. private insurance health care claims to estimate management costs.Helms and Melnikow 19 used a cohort of 98 cervical cancer patients with 1990-1991 health maintenance organization data and 133,058 case controls.Demographic descriptions that would permit a check of the similarity of their cohorts to ours were not available.The mean cost of $30,136 (1996 U.S. dollars) is higher, after adjustment for inflation and currency conversion, than the cost estimated in our study.Using International Classification of Diseases (version 9) codes to identify cervical cancer health care claims, McCrory et al. 18 estimated costs for women 20-64 years of age.After adjustment for inflation and conversion, their mean cost by stage-$17,645 for stage i, $27,069 for stages ii-iii, and $40,280 for stage iv (1997 U.S. dollars)-are higher, but follow the pattern observed in the present study.Lastly, a recent retrospective cost and resource utilization analysis of hpv-related cancers by Insinga et al. 17 calculated a mean mortality-adjusted cost of $29,649 (2003 U.S. dollars), which is higher than our study's mean cost.The Insinga cohort was notably younger (age 60 or older: 10.1% vs. 26%), and the associated resource utilization estimates were different from those in the present study.Cervical cancer patients experienced, on average, 12.0 outpatient visits (95% confidence interval: 9.6 to 14.4 visits) and 0.6 inpatient visits (95% confidence interval: 0.5 to 0.7 visits), but the length of inpatient visits was not presented.The difference in the number of hospital Cost comparisons across international jurisdictions are difficult for a number of reasons.The structure of the U.S. health care system is quite different from that of single-payer systems such as Canada's.Private health insurance in the United States is often employer-based 17 and reflects costs for a working-age population.In 2004, the average age of retirement for U.S. women was 62 years 28 .As stated earlier, the percentage of women more than 60 years of age was significantly higher in the present analysis than in earlier studies.Thus, management costs based on private health insurance might not capture the experience of older women.

Study Strengths
To our knowledge, this Canadian population-based study is the first to report resource use and costs associated with the management of cervical cancer from diagnosis to discharge.Analysis of resource-based patient-level costs has numerous strengths.Real-world patient experience is difficult to quantify in a model because oncologists personalize disease management depending on factors such as patient preference, age, and fertility.Retrospective patient-level costing aims to determine the actual resources used and allows researchers to separately analyze subgroups of patients so as to identify variations in disease management patterns.
Colleagues in Ontario recently published a similar retrospective exercise estimating costs 1 year after diagnosis of various cancers in Ontario 29 .Their analysis suggests that cervical cancer is not among the most costly cancers to treat.Much of the lower cost could be attributable to the fact that many cancers are detected through screening and can be resolved with surgery.It might be reasonable to assume that, in the absence of a major change in the approach to treatment, cervical cancers are less resource-intensive to manage than are cancers at other sites.
Our analysis was also able to capture a comprehensive picture of health care resource utilization in cervical cancer patients.Hospital-based costing exercises often cannot capture the contribution of services delivered outside of hospital-for example, doctor's visits, communitybased pharmacy, and so on.By linking bcca data with provincial ministry data, we were able to reach an estimate that is likely reflective of the main components of cost.Unlike costing analyses that use clinical trial data, our results include data from patients across the province and likely reflect the full breadth of treatment experience in a province of more than 4 million people.That comprehensiveness is also reflected in our screening data: the ccsp is responsible for all cervical cancer screening in the province of British Columbia.
Reimbursement of health professionals, drug purchase costs, administration costs, the role of private insurers, and a number of other factors differ across national and regional lines.As a result, costing exercises are often limited in their transferability outside their local context.However, our analysis provides resource utilization as well as cost estimates, allowing decision-makers in other jurisdictions to use their own unit costs to adapt our findings to their local context.

Limitations
Despite its listed strengths, our study was still based on results from a single Canadian province.The further our findings are removed from that context (that is, to another Canadian province, a publicly-funded health care system in another country, or another country with a different health care structure), the less generalizable they become.By publishing resource utilization in addition to cost, we hope to make our results relevant across a wide variety of decision-making contexts; however, readers should be aware that our costs might not accurately reflect cervical cancer management outside of British Columbia.Nevertheless, we feel that the analysis provides a rigorous estimate of the cost of cervical cancer management and could still be highly relevant across and outside of Canada.
Investigation into hospitalization records revealed that some hospital visits included brachytherapy procedure codes.That discovery was problematic, because the cost of brachytherapy was separately estimated using the number of insertions, suggesting the potential for double-counting.Of all brachytherapy visits, 64% had a corresponding hospitalization record and a cost lower than the expert-derived cost of $3,705 per insertion.We decided that hospitalization records with a brachytherapy procedure should be included, because removing them would have resulted in the exclusion of 36% of insertions and lower-than-expected management costs.It is therefore possible that the brachytherapy cost could be slightly overestimated.We believe that the estimates are still largely representative of the true cost of treatment, but we recognize that a different approach might be needed to provide a more rigorous estimate of the costs associated with brachytherapy.
The retrospective nature of our analysis necessarily means that our study cohort is "dated" with respect to the current standard of practice for cervical cancer management (for example, the advent of biologic agents such as bevacizumab for advanced disease 30 ).Regardless, we feel that our results are highly useful as a "baseline" for use in cost-effectiveness analyses of novel approaches.

CONCLUSIONS
Based on resource utilization determined using a cohort of 563 women diagnosed over a period of 5 years, managing a case of cervical cancer in British Columbia costs the health care system an average of $19,153 (standard error: $3484).Hospitalization represented the largest component of that cost.
In light of emerging technologies and approaches to screening, prevention, and treatment of cervical cancer, the results of the present analysis are intended to be useful for estimating the economic impact and cost-effectiveness of changes to cervical cancer health care policy.

TABLE II
Mean resource use by resource type and clinical subgroup

(1.6) 1.8 (2.5)
a Boldface type signifies that differences in means within this category are statistically significantly different from zero (Kruskal-Wallis H-test, p < 0.05).b Defined as having at least 1 screening appointment between 6 months and 5 years before the cancer diagnosis.Pts = patients; CCSP = Cervical Cancer Screening Program.

TABLE III
Mean survival and cost, overall and by subgroup Vol. 23, Supp. 1, February 2016 © 2016 Multimed Inc. visits in Insinga et al. and in our analysis could be a result of the two-stage regression and case-control method used by Insinga et al. to estimate resource utilization.Findings in the Insinga et al. paper were presented as mean values; our study presents resource utilization values as medians because of the skewed nature of the data.
a With standard error.bDefined as having at least 1 screening appointment between 6 months and 5 years before the cancer diagnosis.Pts = patients; CCSP = Cervical Cancer Screening Program.Current Oncology,

TABLE IV
Mean proportion of total cost by resource type Defined as having at least 1 screening appointment between 6 months and 5 years before the cancer diagnosis.Pts = patients; MSP = Medical Services Plan; CCSP = Cervical Cancer Screening Program.
a Boldface type signifies that differences in means within this category are statistically significantly different from zero (analysis of variance, p < 0.05) b Current Oncology, Vol. 23, Supp. 1, February 2016 © 2016 Multimed Inc.