Background

American Society of Clinical Oncology (ASCO) guidelines recommend the integration of specialist palliative care (SPC) into oncology care [1]. Reviews of randomized trials found that SPC improves patient and family experience, including symptom burden and psychological distress, and increased probability of death in preferred place [2, 3].

While SPC capacity has increased substantially over the last two decades in the US [4–7] and internationally [8, 9], most people with palliative care needs do not receive SPC [10, 11]. In some cases this may reflect patient preferences [12], but other key determinants include insufficient PC capacity, sub-optimal referral rates among oncologists and low awareness of PC among patients [13–15]. There are potential inequities in access to scarce SPC capacity, including by diagnosis, race and ethnicity, socioeconomic position, and urban and rural locations [16–18].

There is a need for better evidence to understand health disparities US [19, 20]. Robust data on how many people with cancer in the US receive SPC, and the characteristics and representativeness of that population, are rare for several reasons. First, people with cancer receive care across different settings without routine data linkage [10]. Second, it is often difficult to determine which people actually received care from palliative care specialists using a single source of secondary data [21].

Rationale and aims

In the context of these evidence needs and data challenges, we compiled a population-representative dataset to understand SPC engagement among people with cancer in the Richmond-Petersburg metropolitan statistical area (Richmond MSA) in Virginia for 2010–2015. Our specific research questions were: what proportion of people with cancer in the Richmond MSA used SPC?; what socio-demographic and clinical characteristics were associated with SPC receipt?; and what is the timing of SPC relative to death and what characteristics are associated with earlier receipt?

Data sources

The Virginia Cancer Registry (VCR) of the Virginia Department of Health (VDH) provided statewide data on May 7, 2018 on adult cancer decedents (> = 21 at time of death) who died 2012–2015. Data elements from the VCR included patient-level identifiers and demographics; and tumor-level diagnosis dates, cancer types, histology, stage at diagnosis, and locality of residence (including census tract) at time of diagnosis.

Virginia Health Information (VHI) provided statewide data on all hospitalizations from 2010–2015 for those cancer decedents on September 12, 2019. Data elements included admission and discharge dates, diagnoses, procedures, charges, insurance type, and hospital identifiers.

The three palliative care programs provided patient-level data on SPC services from 2010–2015 in December 2019. Data elements included patient identifiers and dates of services.

Authors representing the three palliative care providers (D. Buxton, D. Noreika, E. Del Fabbro, L. Yanni, S. Roberts) had access to patient identifiers only for their own programs prior to submitting their data. One author (N. Skoro) had access to patient identifiers in the data from the cancer registry and palliative care provider groups; the data were de-identified prior to further analysis.

Availability of SPC

Information about availability of SPC services was obtained directly from the SPC programs. Within the Richmond MSA, there were 3 distinct medical groups providing SPC services at 6 hospitals. One was an academic medical center providing hospital-PC and clinic-PC; a second group was at Catholic, non-profit system with four hospital locations offering hospital-PC, office-PC and home-PC; and a third group was at a for-profit hospital. All three programs had multi-disciplinary teams; two were certified by the Joint Commission’s hospital-PC certification program [25]. These were the only SPC providers in the Richmond MSA in the study time period. Emails, websites, and phone calls confirmed that other hospitals within or nearby the area did not have SPC services during the years of interest (2010–2015), and that no other medical groups or insurance-based providers were operating in the region during that time. The six hospitals within the MSA that did not offer SPC were for-profit. None of the 5 small, rural hospitals just outside of the MSA offered palliative care services. See Fig 1.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Availability of specialist palliative care within and around the Richmond MSA.

Hospitals within and near the Richmond-Petersburg Metropolitan Statistical Area. Map generated using ArcGIS Pro 3.0.0. Hospital addresses source: Virginia Hospital & Health System Association [26]. SPC = specialty palliative care.

https://doi.org/10.1371/journal.pone.0313732.g001

Data linkage

VCR data were linked to hospitalization data by VHI using hashed (encrypted) social security numbers (SSN). VCR data were linked by the researchers to SPC data using patient name and date of birth. The resulting VCR-VHI-SPC dataset was then linked to US census data based on census tract identifiers at the time of diagnosis.

Inclusion and exclusion criteria

VCR data included cancer registry cases (primary malignant neoplasms) for people who were at least 21 years old at the time of diagnosis and were known to have died between 2012–2015. VCR excluded cases that were diagnosed or treated within the Veterans Health Administration system per VCR requirements. We also excluded people whose only cancer was a non-melanoma skin cancer as that is quite common, usually not fatal, and not typically the basis for referrals to palliative care [27]. Using the combined VCR-VHI data, we included people that were living within the Richmond MSA at the time of cancer diagnosis, and excluded cases that had additional cancers diagnosed outside the MSA. We excluded cases that had hospitalizations outside of the Richmond MSA (beyond the several small, rural hospitals just outside the MSA, none of which had SPC services) because we had palliative care data only from providers within the MSA.

In the dataset provided by the VCR there were 13,260 adult cancer decedents who died at age 21+ in years 2012–2015 who were living in the Richmond MSA at the time cancer diagnosis. We excluded 1,230 people because they either did not have a valid SSN (n = 91) which would have prevented linking them to VHI hospitalization data; or had hospitalizations beyond the 17 hospitals within or nearby the Richmond MSA within 2 years of death (n = 792) which would have prevented our assessment of their PC access or exposure; or had irreconcilable dates from multiple data sources (n = 377) such as admission, discharge dates or initial PC dates after date of death, or incomplete diagnosis dates. The remaining dataset contained 12,030 cancer decedents.

Variables used in this study

For each patient, the data included at least two years of potential hospitalizations and SPC encounters prior to death. Because we were examining factors related to receiving palliative care, the number of hospitalizations as a potential predictor variable was measured prior to initiation of SPC if received, else prior to death.

For the descriptive analyses of what proportion of the cancer population used SPC, our outcome variables included a binary variable of SPC use, and the timing of first SPC encounter prior to death. Predictor variables included age at time of cancer diagnosis, age at time of death, sex, race, cancer type, rurality, and socio-economic status (SES). A rurality index was used, based on the percent of the population living in non-urban areas divided into 4 groups representing 0% rural, 1–50% rural, 51–99% rural, and 100% rural [28]. The socio-economic status variable was derived from 7 data elements from the American Community Survey census data and converted to quintiles [29]. Imputation for SES was performed where only 5 or 6 of those data elements were available; quintiles were ranked from low SES to high SES. Cancers were characterized in two ways: solid versus hematologic, and poor prognosis or not. Cancers with poor prognosis were those that nationally had 40% or fewer surviving for 5 years [30, 31]; or had distant metastases at diagnosis.

Statistical analyses

Frequencies and percentages were used to describe categorical variables, with means (standard deviations) and/or median (interquartile range) used for continuous variables. Univariate association with SPC used Chi-square (categorical variables), or Wilcoxon Rank Sum test (continuous variables). Multivariable analysis of SPC use used backward logistic regression, with variables staying in the model as long as p< = 0.05. Cox proportional hazards regression was used to evaluate characteristics associated with timing of SPC for those who used SPC. Analyses were conducted in SAS [32] and the map (Fig 1) was generated using ArcGIS Pro [33].

Ethics

The Virginia Department of Health (VDH) IRB approved the study on May 2, 2018 (study #40256) and the Virginia Commonwealth University (VCU) IRB exempted this study on January 8, 2019 (HM20015277). Data use agreements were established with VDH, VHI, and the two medical groups outside of VCU: Bon Secours Medical Group and the Apollo Medical Group. The data privacy, compliance and legal officers of each medical group approved the acquisition of data on patients’ SPC use. The data use agreements with those entities as well as Virginia Health Information and Virginia Department of Health forbid the sharing of the datasets with other entities due to privacy concerns. The Virginia Department of Health IRB, and the privacy, compliance and legal offices of the data providers (Bon Secours Medical Group, the Virginia Commonwealth University Health System faculty practice, the Apollo Medical Group, Virginia Health Information, and Virginia Department of Health) waived the need for consent as this study concerned decedents. The study authors report no conflicts of interest.

Patient characteristics

Clinical, demographic, geographic and socio-economic characteristics of the cancer population are provided in the first column of Table 2.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Univariate analyses of SPC use (n = 12,030).

https://doi.org/10.1371/journal.pone.0313732.t002

Use of local hospitals

Hospital admission was evaluated at 24, 18 and 6 months before death. Looking at all 12,030 cancer decedents and the full 24-month period, 10,273 (85.4%) went to a hospital; of them, 8,051 (78.4%) went to a hospital offering SPC. In the 18 months before death, 9,914 (82.4%) had at least one hospitalization, and 7,671 of them (77.4%) went to a hospital offering SPC. In the 6 months before death, 9005 (74.9%) were hospitalized at least once and 6,815 of them (75.7%) went to a hospital offering SPC.

Some patients used more than 1 hospital. Of the 6,614 patients who had at least 1 hospitalization the 30 days prior to PC or death, 344 (5.2%) went to more than 1 hospital. Of the 8,301 patients who had at least 1 hospitalization over the 3 months prior to PC or death, 921 (11.1)% went to more than 1 hospital. Of the 9,005 patients who had at least 1 hospitalization over the 6 months prior to PC or death, 1,333 (14.8%) went to more than 1 hospital.

Use and timing of specialty palliative care

One-quarter of the cancer decedents (n = 2,958, 24.6%) received SPC from one or more sources in the 18 months before death. Of the 7,671 who went to a hospital offering SPC within 18 months before death, 2,671 (34.8%) received SPC; of the other 4,359 who did not go to a hospital offering SPC, 287 (6.6%) received SPC in clinic or home settings. While most SPC recipients (97.5%) received SPC from a single medical group, 73 received SPC from two medical groups and 1 person received SPC from all 3.

Of the 9,072 persons who did not use palliative care, 3,877 (42.7%) went only to hospitals that did not offer SPC; 3,517 (38.8%) went to hospitals that offered SPC but did not use it; and the remaining 1,678 (18.5%) did not go to any hospital (and did not use community-based SPC).

From first cancer diagnosis to first SPC encounter, a mean of 4.5 years passed (SD 6.5), with a median of 1.4 years (IQR 5.9). From first SPC encounter to death, the mean was 123.9 days (SD 243) with a median of 30 days (IQR 107). From first SPC encounter to death, 1493 (50.5%) were in the final 30 days of life, including 768 (26.0%) in the final week of life, 450 (15.2%) in the final 3 days of life, and 126 (4.3%) had a first SPC encounter on the day they died.

Characteristics associated with SPC use and timing

Most of the characteristics of the population that we evaluated were significantly associated with whether SPC was used, as shown in Table 2. The rate of SPC use increased over the years, from around 22% for those dying in 2012 or 2013, and increasing to 28% by 2015.

As shown in Table 3, the following characteristics were associated with using SPC in the multivariate analysis: being in an SES quintile other than the lowest; being younger; being Black; having a solid (versus hematological) cancer; having a shorter survival with cancer; dying in the latter two years of the study; being from an area of low or complete rurality; having a hospital admission in the final 60 days prior to initiation of PC or death; having more days in hospital; and living within 15 miles of a hospital offering SPC. The AUC statistic was 0.749.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Multivariate analysis of SPC use (n = 11,689).

https://doi.org/10.1371/journal.pone.0313732.t003

Two secondary regressions were performed: one model limited to patients admitted to hospitals offering SPC (to control for access to SPC), and one model limited patients with poor-prognosis cancers or stage (which are commonly targeted in SPC intervention studies). Similar AUCs were achieved and certain variables were significant predictors across models: younger age, being Black, more recent year, shorter distance to SPC hospital, rurality group, and more days in hospital in final 6 months. See S1 and S2 Tables.

Univariate analyses and Cox proportional hazards regression were also conducted to determine characteristics associated with the timing of SPC relative to death, for those who used SPC (n = 2,792). Few variables were statistically significant in the multivariate analysis, as shown in Table 4. The characteristic associated with longest time from SPC to death was having zero hospitalizations in the six months prior. A number of variables that were associated with SPC use, such as race, cancer type, rurality and SES, were not significantly associated with the timeliness of SPC among those who used it.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Characteristics associated with days from SPC to death, multivariate analysis.

https://doi.org/10.1371/journal.pone.0313732.t004

Methodological strengths

The current study used population-based data sets appropriate for determining population-level use of SPC. The starting place was population data in the form of the state cancer registry. This was then linked to another population data source, state-wide hospital discharges. Finally, institutional data on SPC use was acquired from SPC providers in the MSA.

This approach had several strengths. The use of population-based data avoids sampling bias. Not all patients were hospitalized, and because early palliative care may reduce the need for subsequent hospitalizations [53–55], it was critical to start with all cancer decedents regardless of hospital use. Second, we used statewide cancer registry and hospital discharge data to exclude people who were diagnosed or hospitalized outside of the metropolitan area, thus excluding people for whom we could not evaluate SPC use. Most importantly, we did not rely on the ICD9/10 codes for “palliative care encounter” (V66.7 / Z51.5) which are not suitable for research on SPC [21].

To our knowledge, this approach has not been used in prior studies of SPC in the US. It is a replicable method that would work well in other small- and medium-sized metropolitan statistical areas. The challenge of using this approach in larger areas would be in securing data use agreements with larger numbers of SPC providers.