Abstract
Guideline-concordant treatment (GCT) of lung cancer has been observed to vary across geographic regions over the years. However, there is little evidence as to what extent this variation is explained by differences in patients’ clinical characteristics versus contextual factors, including socioeconomic inequalities.
This study evaluated the independent effects of individual- and area-level risk factors on geographic and temporal variation in receipt of GCT among patients with lung cancer. Receipt of GCT was defined on the basis of the National Comprehensive Cancer Network guidelines. We used Bayesian spatial-temporal multilevel models to combine individual and areal predictors and outcomes while accounting for geographically structured and unstructured correlation and linear and nonlinear trends.
Our study included 4,854 non–small cell lung cancer (NSCLC) and small cell lung cancer (SCLC) cases, reported to the Victorian Lung Cancer Registry between 2011 and 2018. Area-level data comprised socioeconomic disadvantage and remoteness data at the local government area level in Victoria, Australia. Around 60.36% of patients received GCT, and the rates varied across geographic areas over time. This variation was mainly associated with poor performance status, advanced clinical stages, NSCLC types, public hospital insurance, area-level deprivation, and comorbidities.
This study highlights the need to address disparities in receipt of GCT among patients with lung cancer with poor performance status, NSCLC, advanced clinical stage, stage I–III SCLC, stage III NSCLC, public hospital insurance, and comorbidities, and living in socioeconomically disadvantaged areas.
Two-year mortality outcomes significantly improved with GCT. Interventions aimed at reducing these inequalities could help to improve lung cancer outcomes.
Introduction
Lung cancer is the leading cause of cancer-related deaths in Australia, with a lower 5-year relative survival rate (17.4%) than other cancers combined (68.9%; refs. 1–3). Clinical practice guidelines inform clinical management, aimed at reducing variation and improving quality of care, and these are based on expert consensus and available evidence (4).
The National Comprehensive Cancer Network (NCCN; refs. 5, 6) has provided clinical practice guidelines for the treatment of small cell lung cancer (SCLC) and non–small cell lung cancer (NSCLC). Adherence to these guidelines was associated with survival benefits (7–9). Despite the established clinical guidelines (5, 6), the extent to which patients with SCLC and NSCLC received guideline-concordant treatment (GCT) is unknown in Victoria, Australia. Studies of clinical practice patterns in Australia have reported variations in the treatment of lung cancer by geographic areas over the years (10). However, it is unclear how much of this variation could be explained by the influence of patients’ clinical characteristics versus hospital factors and contextual factors such as socioeconomic inequalities. Identifying and quantifying the magnitude of geographic disparities in relation to adherence to GCT could be an essential strategy toward improving the overall quality of lung cancer treatment and long-term survival outcomes.
The use of traditional nonspatial models for the evaluation of geographic data can be misleading because individuals in the same area or year tend to have similar characteristics (spatial-temporal correlation) and these need to be accounted for in the analysis (11–13). Although multilevel models allow for the simultaneous estimation of individual- and area-level effects, they cannot account for spatial associations between geographic areas (14). They may be limited by data sparsity in each area for rare outcomes (14). In contrast, spatial models can incorporate spatial dependence among neighboring areas through the inclusion of the conditional autoregressive (CAR) prior while borrowing information from adjacent areas to overcome data sparseness for rare outcomes (15). They allow stable small area RR estimates, particularly for regional areas with a smaller population (15). To date, there have been no studies that considered the spatial and temporal variations within the multilevel modeling framework to evaluate lung cancer outcomes, including receipt of GCT.
This study aimed to evaluate the independent effects of individual- and area-level risk factors on geographic and temporal variation in the receipt of GCT among patients with lung cancer. Within the hierarchical-related regression (HRR) approach, this study used Bayesian spatial-temporal multilevel models to combine individual and areal predictors with outcomes. It accounts for geographically structured and unstructured correlation and linear and nonlinear trends in one comprehensive and parsimonious model.
Materials and Methods
Data
Individual level
This study analyzed deidentified patients with lung cancer from the Victorian Lung Cancer Registry (VLCR), recruited between 2011 and 2018. This study was approved by the Monash University (Melbourne Victoria, Australia) Human Research Institutional Review Board (project ID: 22529) and complies with the Declaration of Helsinki. Written informed consent was not obtained as the patients’ participation in VLCR was based on an opt-out process, and this was detailed to them in the patient information sheet (16). VLCR collects data from 19 Victorian hospitals and covered around 85% of Victorian lung cancer notifications in 2018 (16). VLCR recruited patients if they were at least 18 years old and presented with incident lung cancer cases based on either a clinical or pathologic diagnosis. It excluded cases with secondary lung cancer or mesothelioma. Data on age, sex, histologic types, comorbidities, smoking, stage, Eastern Cooperative Oncology Group (ECOG) performance status (PS), date of diagnosis, types and dates of treatments, notifying hospital insurance status (public and private), and multi-disciplinary team meetings (MDM) were obtained. For each clinical subgroup of lung cancer type and stage, GCT, the minimal first-course treatment patients should receive, was determined on the basis of the NCCN guidelines (refs. 5, 6; Table 1). ECOG scores quantify the functional status of patients with cancer with a 5-point scale, from being able to carry out normal activities (ECOG, 0) to being bed bound (ECOG, 4; ref. 17). For each referring hospital, we included hospital volume, a well-established indicator of quality of care (18), derived by calculating the number of patients treated at a particular hospital and categorized into tertiles.
Disease . | Stage subgroup . | GCTa . |
---|---|---|
NSCLC | Localized (stages I–II) | Surgery ± additional treatments and/or SBRT ± additional treatments |
Locally advanced (stage III) | Surgery + chemotherapy ± additional treatments and/or radiotherapy (any regimen) + chemotherapy ± additional treatments | |
Advanced (stage IV) | Chemotherapy ± additional treatments | |
SCLC | Limited (stage I–III) | Surgery + chemotherapy ± additional treatments and/or radiotherapy (any regimen) + chemotherapy ± additional treatments |
Extensive (stage IV) | Chemotherapy ± additional treatments |
Disease . | Stage subgroup . | GCTa . |
---|---|---|
NSCLC | Localized (stages I–II) | Surgery ± additional treatments and/or SBRT ± additional treatments |
Locally advanced (stage III) | Surgery + chemotherapy ± additional treatments and/or radiotherapy (any regimen) + chemotherapy ± additional treatments | |
Advanced (stage IV) | Chemotherapy ± additional treatments | |
SCLC | Limited (stage I–III) | Surgery + chemotherapy ± additional treatments and/or radiotherapy (any regimen) + chemotherapy ± additional treatments |
Extensive (stage IV) | Chemotherapy ± additional treatments |
Abbreviation: SBRT, stereotactic body radiotherapy (thoracic radiotherapy with a dose of ≥45 Gy in ≤5 fractions).
aThe minimal treatment patients should receive according to the NCCN guidelines.
Area level
Patients were assigned to local government areas (LGA) on the basis of their residential address at diagnosis. LGA was selected as the spatial unit of analysis due to available census population data from the Australia Bureau of Statistics (ABS; ref. 19) and has been analyzed as an evidence-based tool to target policy interventions effectively (20). Area-level data from ABS 2016 census comprised of socioeconomic disadvantage and remoteness data at the LGA level (79 LGAs) in Victoria, Australia (19, 21). Index of relative socioeconomic disadvantage (IRSD) was used as a surrogate indicator of area-level socioeconomic status (SES; ref. 22). An area with a low IRSD score indicated a high proportion of relatively disadvantaged people in LGA (22). According to the ABS classification, we categorized IRSD into quartiles (22). We used six categories for remoteness: (i) “major city,” (ii) “major city and inner regional,” (iii) “inner regional,” (iv) “inner regional and outer regional,” (v) “outer regional,” and (vi) “outer regional and remote” (21). Categories 1 and 2 correspond to metropolitan/urban areas in Greater Melbourne, and categories 3–6 correspond to regional/rural areas outside Greater Melbourne (21).
Statistical analysis
The statistical models were derived from the HRR model framework (Jackson and colleagues, 2008 and Jonker and colleagues 2015; refs. 23, 24). The HRR framework incorporates both individual-level and area-level outcomes models and simultaneously integrates individual- and area-level predictors and outcomes data within a single, parsimonious framework (Supplementary Materials and Methods).
We modified the HRR model to incorporate spatial dependency by specifying a Besag, York, and Mollié (BYM) model and included both linear and quadratic temporal trends. The BYM model empirically partitions geographic variation into spatially structured and unstructured components. Spatially structured variance component (CAR prior) smooths risk estimates toward the local mean of contiguous area units, and the unstructured variance component (specified via normal distribution) smooths risk estimates toward the global mean of the study unit (25). The Bayesian spatial-temporal model with CAR in the spatial random effect and linear and quadratic terms in the temporal random effects has shown to account for the evolution of disease outcomes across areas over time simultaneously (26, 27).
Individual-level outcomes model
Area-level outcomes model
The traditional ecological study directly models area-level outcomes in terms of area-level mean exposures via a logistic or Poisson regression model. But, the area-level mean of each covariate limits to estimate the separate effects of the individual- or area-level covariates. The corresponding OR is a biased estimate of the individual-level effect and can only allow for area-level and not individual-level inference. Thus, ecological bias can occur due to a within-area variability in both exposures of interest and confounders (28). Our model sought to overcome this limitation.
Random area-level effects such as uj and vj are specified via a BYM CAR prior, a combination of unstructured and spatially structured error terms to allow for spatial correlation in the error term. These components indicate the extent to which geographic variation is due to spatially structured and unstructured factors. Along with the random area effect, we included a random effect coefficient for the area-level IRSD and remoteness. Linear tt and quadratic tt2 time effects were specified to capture linear and nonlinear temporal variations in the log-odds of GCT. The models were estimated in a Bayesian framework using weakly informative priors, which are useful to regularize sparse data settings in HRR applications. We assigned normal (mean = 0 and precision = 0.725) priors to fixed effects individual-level covariates and random effects area-level covariates, which represent the 95% prior belief that the true OR ranges between 1/10 and 10. MultiBUGS software was used to estimate the parameters using iterative Monte Carlo Markov (MCMC) sampling techniques (29).
We performed direct internal standardization using the Victorian census population data as the reference population and calculated the age- and sex-standardized RR for each LGA and year. We conducted χ2 tests to determine the unadjusted associations between categorical variables and GCT. We fitted several different models: unadjusted models (model A, age and gender standardized without a spatial BYM model and temporal random effects), unadjusted models (model B, age and gender standardized with a spatial BYM model and temporal random effects and each covariate in turn), and adjusted models (model C, age and gender standardized with a spatial BYM model, temporal random effects, and covariates).
Model building and selection were based on the deviance information criterion (DIC). Lower DIC values represent a better fit and importance of a covariate. On the basis of the univariate results, we sequentially added covariates to examine whether the covariates were statistically significant, and their inclusion improved the model fit based on DIC. We considered the model selection criterion (30) to suggest models within 1–2 DIC units of the best model (lowest DIC) as strongly supported, 3–7 as having less support, >7 as no support. OR and 95% credible interval (95% Crl) estimates for each parameter were summarized from posterior distributions of two MCMC chains initialized using overdispersed starting values. We ran a total of 50,000 iterations with first 10,000 discarded to reduce autocorrelation and improve convergence. Autocorrelation plots and the Gelman–Rubin diagnostics were used to confirm the convergence of MCMC chains (31, 32).
We mapped the age- and sex-standardized RR by dividing the corresponding age- and sex-stratified observed and expected counts within the particular LGA and year.
Results
Individual-level data included 4,854 cases reported to the VLCR between 2011 and 2018 after excluding cases with other lung cancer types (not NSCLC or SCLC) and unknown clinical stages and invalid residential addresses.
Patterns of treatment and guideline concordance per stage grouping are presented for patients with NSCLC (n = 4,467) and SCLC (n = 387; Table 2). Among localized NSCLC, surgery only was the most common GCT (49.7%) followed by surgery with chemotherapy (12.8%) and stereotactic body radiotherapy (SBRT; 7.68%). Patients with locally advanced NSCLC and limited SCLC received the combination radiotherapy and chemotherapy more commonly as the GCT than the combination surgery and chemotherapy with or without radiotherapy. Conventional radiotherapy with chemotherapy was the most commonly delivered GCT among advanced NSCLC, whereas chemotherapy only was the most common GCT among extensive SCLC. Localized and locally advanced NSCLC received less-intensive-than-recommended treatment, most commonly with conventional radiotherapy alone. However, a higher proportion of patients in the advanced NSCLC and extensive SCLC groups received no treatment, compared with conventional radiotherapy only. Among limited SCLC, chemotherapy only was the most common less-intensive-than-recommended treatment, followed by conventional radiotherapy alone and no treatment.
Clinical stage subgroups . | n (%) . |
---|---|
Localized NSCLC (stages I–II) | 1,016 |
GCTa | 756 (74.41%) |
Surgery only | 505 (49.7%) |
Surgery + chemotherapy | 130 (12.8%) |
SBRT only | 78 (7.68%) |
Surgery + conventional radiotherapyb + chemotherapy | 25 (2.46%) |
Surgery + conventional radiotherapyb | 18 (1.77%) |
Less intensive treatment than recommended | 260 (25.59%) |
Chemotherapy only | 22 (2.17%) |
Conventional radiotherapyb + chemotherapy | 78 (7.68%) |
Conventional radiotherapyb only | 101 (9.94%) |
No treatment | 59 (5.81%) |
Locally advanced NSCLC (stage III) | 1,228 |
GCTa | 583 (47.48%) |
Surgery + chemotherapy | 96 (7.82%) |
Conventional radiotherapyb + chemotherapy | 445 (36.24%) |
Surgery + conventional radiotherapyb + chemotherapy | 42 (3.42%) |
Less intensive treatment than recommended | 645 (52.52%) |
Chemotherapy only | 110 (8.96%) |
Conventional radiotherapyb only | 217 (17.67%) |
Surgery only | 181 (14.74%) |
Surgery + conventional radiotherapyb | 11 (0.9%) |
No treatment | 126 (10.26%) |
Advanced NSCLC (stage IV) | 2,223 |
GCTa | 1,291 (58.07%) |
Conventional radiotherapyb + chemotherapy | 632 (28.43%) |
Chemotherapy only | 611 (27.49%) |
Conventional radiotherapyb + chemotherapy + surgery | 15 (0.67%) |
Chemotherapy + surgery | 33 (1.48%) |
Less intensive treatment than recommended | 932 (41.93%) |
Conventional radiotherapyb only | 445 (20.02%) |
No treatment | 487 (21.91%) |
Limited SCLC (stage I–III) | 141 |
GCTa | 89 (63.12%) |
Surgery + chemotherapy | 2 (1.42%) |
Conventional radiotherapyb + chemotherapy | 82 (58.16%) |
Surgery + conventional radiotherapyb+ chemotherapy | 5 (3.55%) |
Less intensive treatment than recommended | 52 (36.88%) |
Conventional radiotherapyb only | 11 (7.8%) |
Chemotherapy only | 28 (19.86%) |
Surgery only | 6 (4.26%) |
No treatment | 7 (4.96%) |
Extensive SCLC (stage IV) | 246 |
GCTa | 211 (85.77%) |
Conventional radiotherapyb + chemotherapy | 70 (28.46%) |
Chemotherapy only | 141 (57.32%) |
Less intensive treatment than recommended | 35 (14.23%) |
Conventional radiotherapyb only | 7 (2.85%) |
No treatment | 28 (11.38%) |
Clinical stage subgroups . | n (%) . |
---|---|
Localized NSCLC (stages I–II) | 1,016 |
GCTa | 756 (74.41%) |
Surgery only | 505 (49.7%) |
Surgery + chemotherapy | 130 (12.8%) |
SBRT only | 78 (7.68%) |
Surgery + conventional radiotherapyb + chemotherapy | 25 (2.46%) |
Surgery + conventional radiotherapyb | 18 (1.77%) |
Less intensive treatment than recommended | 260 (25.59%) |
Chemotherapy only | 22 (2.17%) |
Conventional radiotherapyb + chemotherapy | 78 (7.68%) |
Conventional radiotherapyb only | 101 (9.94%) |
No treatment | 59 (5.81%) |
Locally advanced NSCLC (stage III) | 1,228 |
GCTa | 583 (47.48%) |
Surgery + chemotherapy | 96 (7.82%) |
Conventional radiotherapyb + chemotherapy | 445 (36.24%) |
Surgery + conventional radiotherapyb + chemotherapy | 42 (3.42%) |
Less intensive treatment than recommended | 645 (52.52%) |
Chemotherapy only | 110 (8.96%) |
Conventional radiotherapyb only | 217 (17.67%) |
Surgery only | 181 (14.74%) |
Surgery + conventional radiotherapyb | 11 (0.9%) |
No treatment | 126 (10.26%) |
Advanced NSCLC (stage IV) | 2,223 |
GCTa | 1,291 (58.07%) |
Conventional radiotherapyb + chemotherapy | 632 (28.43%) |
Chemotherapy only | 611 (27.49%) |
Conventional radiotherapyb + chemotherapy + surgery | 15 (0.67%) |
Chemotherapy + surgery | 33 (1.48%) |
Less intensive treatment than recommended | 932 (41.93%) |
Conventional radiotherapyb only | 445 (20.02%) |
No treatment | 487 (21.91%) |
Limited SCLC (stage I–III) | 141 |
GCTa | 89 (63.12%) |
Surgery + chemotherapy | 2 (1.42%) |
Conventional radiotherapyb + chemotherapy | 82 (58.16%) |
Surgery + conventional radiotherapyb+ chemotherapy | 5 (3.55%) |
Less intensive treatment than recommended | 52 (36.88%) |
Conventional radiotherapyb only | 11 (7.8%) |
Chemotherapy only | 28 (19.86%) |
Surgery only | 6 (4.26%) |
No treatment | 7 (4.96%) |
Extensive SCLC (stage IV) | 246 |
GCTa | 211 (85.77%) |
Conventional radiotherapyb + chemotherapy | 70 (28.46%) |
Chemotherapy only | 141 (57.32%) |
Less intensive treatment than recommended | 35 (14.23%) |
Conventional radiotherapyb only | 7 (2.85%) |
No treatment | 28 (11.38%) |
Abbreviation: SBRT, stereotactic body radiotherapy (thoracic radiotherapy with a dose of ≥45 Gy in ≤5 fractions).
aThe minimal treatment patients should receive according to the NCCN guidelines.
bAll radiotherapy except SBRT.
Table 3 provides the characteristics of the included patients. Around 60.36% received GCT. More than 80.4% of the patients were older than 60 years of age, and 57.27% were men. Patients with advanced clinical stage, NSCLC histology, comorbidities, smoking, poor ECOG PS (≥1), and those notified at public hospitals and higher volume hospitals were less likely to receive GCT. Conversely, patients resident in least disadvantaged areas were more likely to receive GCT. The proportion of patients with SCLC receiving GCT increased by clinical stages, but decreased with clinical stages among NSCLC. Rates of GCT also increased significantly over the years.
. | Treatment guideline concordancea . | . | . | |
---|---|---|---|---|
. | Concordant . | Discordant . | . | . |
. | n (%) . | n (%) . | Total . | Pb . |
2,930 (60.36%) | 1,924 (39.64%) | 4,854 | ||
Age group | <0.001c | |||
<60 | 705 (24.06%) | 246 (12.79%) | 951 | |
60–69 | 1,005 (34.3%) | 487 (25.31%) | 1,492 | |
70–79 | 964 (32.9%) | 695 (36.12%) | 1,659 | |
80+ | 256 (8.74%) | 496 (25.78%) | 752 | |
Gender | <0.001c | |||
Male | 1,623 (55.39%) | 1,157 (60.14%) | 2,780 | |
Female | 1,307 (44.61%) | 767 (39.86%) | 2,074 | |
Lung cancer types | <0.001c | |||
SCLC | 300 (10.24%) | 87 (4.52%) | 387 | |
NSCLC | 2,630 (89.76%) | 1,837 (95.48%) | 4,467 | |
Stage | <0.001c | |||
Stage I | 543 (18.53%) | 82 (4.26%) | 625 | |
Stage II | 236 (8.05%) | 196 (10.19%) | 432 | |
Stage III | 649 (22.15%) | 679 (35.29%) | 1328 | |
Stage IV | 1,502 (51.26%) | 967 (50.26%) | 2,469 | |
Lung cancer types and stage | <0.001c | |||
SCLC | ||||
Stage IV | 211 (7.2%) | 35 (1.82%) | 246 | |
Stage I–III | 89 (3.04%) | 52 (2.7%) | 141 | |
NSCLC | ||||
Stage IV | 1,291 (44.06%) | 932 (48.44%) | 2,223 | |
Stage III | 583 (19.9%) | 645 (33.52%) | 1,228 | |
Stage I–II | 756 (25.8%) | 260 (13.51%) | 1,016 | |
Comorbidities | 1,320 (45.05%) | 1,000 (51.98%) | 2,320 | <0.001c |
Smoking | <0.001c | |||
Never | 401 (13.69%) | 167 (8.68%) | 568 | |
Ever | 2,426 (82.8%) | 1,676 (87.11%) | 4,102 | |
Not stated | 103 (3.52%) | 81 (4.21%) | 184 | |
ECOG PS | <0.001c | |||
Excellent 0d | 900 (30.72%) | 296 (15.38%) | 1,196 | |
Good 1e | 935 (31.91%) | 547 (28.43%) | 1,482 | |
Moderate 2f | 194 (6.62%) | 265 (13.77%) | 459 | |
Poor 3+4g | 56 (1.91%) | 183 (9.51%) | 239 | |
Not stated | 845 (28.84%) | 633 (32.9%) | 1,478 | |
Hospital insurance status | <0.001c | |||
Private hospital | 547 (18.67%) | 230 (11.95%) | 777 | |
Public hospital | 2,383 (81.33%) | 1,694 (88.05%) | 4,077 | |
Hospital volume | <0.001c | |||
1st quartile | 1,100 (37.54%) | 572 (29.73%) | 1,672 | |
2nd quartile | 1,048 (35.77%) | 758 (39.4%) | 1,806 | |
3rd quartile (highest) | 782 (26.69%) | 594 (30.87%) | 1,376 | |
MDMs | 2,021 (68.98%) | 1,309 (68.04%) | 3,330 | 0.49 |
Area-level socioeconomic disadvantage | 0.001c | |||
1st quartile (most disadvantaged) | 676 (23.07%) | 546 (28.38%) | 1,222 | |
2 | 757 (25.84%) | 472 (24.53%) | 1,229 | |
3 | 768 (26.21%) | 462 (24.01%) | 1,230 | |
4th quartile (least disadvantaged) | 729 (24.88%) | 444 (23.08%) | 1,173 | |
Area-level remoteness | 0.173 | |||
Major cities | 1,324 (45.19%) | 867 (45.06%) | 2,191 | |
Major city and inner regional | 900 (30.72%) | 607 (31.55%) | 1,507 | |
Inner regional | 412 (14.06%) | 227 (11.8%) | 639 | |
Inner regional and outer regional | 195 (6.66%) | 143 (7.43%) | 338 | |
Outer regional | 38 (1.3%) | 30 (1.56%) | 68 | |
Outer regional and remote | 61 (2.08%) | 50 (2.6%) | 111 | |
Year of diagnosis | <0.001c | |||
2011 | 20 (1.04%) | 20 (0.68%) | 40 | |
2012 | 99 (5.15%) | 76 (2.59%) | 175 | |
2013 | 274 (14.24%) | 216 (7.37%) | 490 | |
2014 | 336 (17.46%) | 245 (8.36%) | 581 | |
2015 | 361 (18.76%) | 265 (9.04%) | 626 | |
2016 | 513 (26.66%) | 360 (12.29%) | 873 | |
2017 | 645 (33.52%) | 392 (13.38%) | 1,037 | |
2018 | 682 (35.45%) | 350 (11.95%) | 1,032 |
. | Treatment guideline concordancea . | . | . | |
---|---|---|---|---|
. | Concordant . | Discordant . | . | . |
. | n (%) . | n (%) . | Total . | Pb . |
2,930 (60.36%) | 1,924 (39.64%) | 4,854 | ||
Age group | <0.001c | |||
<60 | 705 (24.06%) | 246 (12.79%) | 951 | |
60–69 | 1,005 (34.3%) | 487 (25.31%) | 1,492 | |
70–79 | 964 (32.9%) | 695 (36.12%) | 1,659 | |
80+ | 256 (8.74%) | 496 (25.78%) | 752 | |
Gender | <0.001c | |||
Male | 1,623 (55.39%) | 1,157 (60.14%) | 2,780 | |
Female | 1,307 (44.61%) | 767 (39.86%) | 2,074 | |
Lung cancer types | <0.001c | |||
SCLC | 300 (10.24%) | 87 (4.52%) | 387 | |
NSCLC | 2,630 (89.76%) | 1,837 (95.48%) | 4,467 | |
Stage | <0.001c | |||
Stage I | 543 (18.53%) | 82 (4.26%) | 625 | |
Stage II | 236 (8.05%) | 196 (10.19%) | 432 | |
Stage III | 649 (22.15%) | 679 (35.29%) | 1328 | |
Stage IV | 1,502 (51.26%) | 967 (50.26%) | 2,469 | |
Lung cancer types and stage | <0.001c | |||
SCLC | ||||
Stage IV | 211 (7.2%) | 35 (1.82%) | 246 | |
Stage I–III | 89 (3.04%) | 52 (2.7%) | 141 | |
NSCLC | ||||
Stage IV | 1,291 (44.06%) | 932 (48.44%) | 2,223 | |
Stage III | 583 (19.9%) | 645 (33.52%) | 1,228 | |
Stage I–II | 756 (25.8%) | 260 (13.51%) | 1,016 | |
Comorbidities | 1,320 (45.05%) | 1,000 (51.98%) | 2,320 | <0.001c |
Smoking | <0.001c | |||
Never | 401 (13.69%) | 167 (8.68%) | 568 | |
Ever | 2,426 (82.8%) | 1,676 (87.11%) | 4,102 | |
Not stated | 103 (3.52%) | 81 (4.21%) | 184 | |
ECOG PS | <0.001c | |||
Excellent 0d | 900 (30.72%) | 296 (15.38%) | 1,196 | |
Good 1e | 935 (31.91%) | 547 (28.43%) | 1,482 | |
Moderate 2f | 194 (6.62%) | 265 (13.77%) | 459 | |
Poor 3+4g | 56 (1.91%) | 183 (9.51%) | 239 | |
Not stated | 845 (28.84%) | 633 (32.9%) | 1,478 | |
Hospital insurance status | <0.001c | |||
Private hospital | 547 (18.67%) | 230 (11.95%) | 777 | |
Public hospital | 2,383 (81.33%) | 1,694 (88.05%) | 4,077 | |
Hospital volume | <0.001c | |||
1st quartile | 1,100 (37.54%) | 572 (29.73%) | 1,672 | |
2nd quartile | 1,048 (35.77%) | 758 (39.4%) | 1,806 | |
3rd quartile (highest) | 782 (26.69%) | 594 (30.87%) | 1,376 | |
MDMs | 2,021 (68.98%) | 1,309 (68.04%) | 3,330 | 0.49 |
Area-level socioeconomic disadvantage | 0.001c | |||
1st quartile (most disadvantaged) | 676 (23.07%) | 546 (28.38%) | 1,222 | |
2 | 757 (25.84%) | 472 (24.53%) | 1,229 | |
3 | 768 (26.21%) | 462 (24.01%) | 1,230 | |
4th quartile (least disadvantaged) | 729 (24.88%) | 444 (23.08%) | 1,173 | |
Area-level remoteness | 0.173 | |||
Major cities | 1,324 (45.19%) | 867 (45.06%) | 2,191 | |
Major city and inner regional | 900 (30.72%) | 607 (31.55%) | 1,507 | |
Inner regional | 412 (14.06%) | 227 (11.8%) | 639 | |
Inner regional and outer regional | 195 (6.66%) | 143 (7.43%) | 338 | |
Outer regional | 38 (1.3%) | 30 (1.56%) | 68 | |
Outer regional and remote | 61 (2.08%) | 50 (2.6%) | 111 | |
Year of diagnosis | <0.001c | |||
2011 | 20 (1.04%) | 20 (0.68%) | 40 | |
2012 | 99 (5.15%) | 76 (2.59%) | 175 | |
2013 | 274 (14.24%) | 216 (7.37%) | 490 | |
2014 | 336 (17.46%) | 245 (8.36%) | 581 | |
2015 | 361 (18.76%) | 265 (9.04%) | 626 | |
2016 | 513 (26.66%) | 360 (12.29%) | 873 | |
2017 | 645 (33.52%) | 392 (13.38%) | 1,037 | |
2018 | 682 (35.45%) | 350 (11.95%) | 1,032 |
aThe minimal treatment patients should receive according to the NCCN guidelines.
bχ2 test.
cStatistically significant (P < 0.05).
dFully active (able to carry out all normal activity without restriction).
eRestricted in physically strenuous activity (ambulatory and able to carry out light work).
fAmbulatory and capable of all self-care (unable to carry out any work activities).
gCapable of only limited self-care (confined to bed or chair more than 50% of waking hours)/completely disabled (not able to self-care, totally confined to bed or chair).
Table 4 displays unadjusted and adjusted OR estimates. Likelihood of receiving GCT decreased significantly among patients with NSCLC, advanced clinical stages, stage I–III SCLC, stage III NSCLC, poor ECOG PS (≥1), public hospital insurance, and increasing hospital volume.
. | . | Model B . | . | Model C . | Model D . |
---|---|---|---|---|---|
. | Model A . | Unadjusted OR . | . | Adjusted OR . | Adjusted OR . |
. | (No covariate)a . | (95% Crl)a . | DIC . | (95% Crl)a . | (95% Crl)a . |
Individual level | |||||
Lung cancer types | 6,864 | ||||
SCLC | Reference | Reference | Reference | ||
NSCLC | 0.45 (0.37–0.55)b | 0.46(0.38–0.56)b | 0.52(0.43–0.63)b | ||
Stage | 6,668 | ||||
Stage I | Reference | Reference | Reference | ||
Stage II | 0.22 (0.16–0.29)b | 0.21 (0.15–0.28)b | 0.21 (0.15–0.28)b | ||
Stage III | 0.19 (0.15–0.24)b | 0.19 (0.14–0.24)b | 0.18 (0.14–0.24)b | ||
Stage IV | 0.31 (0.24–0.39)b | 0.36 (0.28–0.46)b | 0.34 (0.27–0.44)b | ||
Lung cancer types and stage (interaction) | 6,683 | ||||
SCLC | |||||
Stage IV | Reference | ||||
Stage I–IIIc | 0.45 (0.28–0.71)b,c | ||||
NSCLC | |||||
Stage IV | Reference | ||||
Stage IIId | 0.67 (0.58–0.78)b,d | ||||
Stage I–IId | 2.01 (1.7–2.39)b,d | ||||
Comorbidities | 0.97 (0.87–1.08) | 6,916 | 0.73 (0.64–0.83)b | ||
Smoking | 6,900 | ||||
Never | Reference | ||||
Ever | 0.74 (0.54–1) | ||||
Not stated | 1 (0.1–9.96) | ||||
ECOG PS | 6,564 | ||||
Excellent 0e | Reference | Reference | Reference | ||
Good 1f | 0.51 (0.43–0.61)b | 0.49 (0.41–0.59)b | 0.51 (0.42–0.61)b | ||
Moderate 2g | 0.2 (0.16–0.25)b | 0.18 (0.14–0.23)b | 0.19 (0.15–0.24)b | ||
Poor 3+4h | 0.08 (0.06–0.11)b | 0.07 (0.05–0.09)b | 0.07 (0.05–0.1)b | ||
Not stated | 0.43 (0.36–0.51)b | 0.36 (0.3–0.43)b | 0.37 (0.31–0.45)b | ||
Hospital insurance status | 6,863 | ||||
Private hospital | Reference | Reference | Reference | ||
Public hospital | 0.63 (0.55–0.73)b | 0.68 (0.57–0.83)b | 0.7 (0.57–0.84)b | ||
Hospital volume | 6,899 | ||||
1st quartile | Reference | ||||
2nd quartile | 0.88 (0.74–1.04) | ||||
3rd quartile | 0.82 (0.69–0.98)b | ||||
MDMs | 1.01 (0.9–1.13) | 6,918 | |||
Area level | |||||
Socioeconomic disadvantage | 6,906 | ||||
1 (most disadvantaged) | Reference | Reference | Reference | ||
2 | 1.09 (0.84–1.41) | 1.44 (1.09–1.95)b | 1.43 (1.09–1.93)b | ||
3 | 1.28 (0.98–1.7) | 1.51 (1.14–2.03)b | 1.48 (1.12–2.01)b | ||
4th quartile (least disadvantaged) | 1.24 (0.93–1.7) | 1.64 (1.19–2.31)b | 1.61 (1.14–2.37)b | ||
Remoteness | 6,840 | ||||
Major cities | Reference | ||||
Major city and inner regional | 1.09 (0.84–1.4) | ||||
Inner regional | 1.36 (0.95–1.98) | ||||
Inner regional and outer regional | 1.02 (0.67–1.57) | ||||
Outer regional | 1.2 (0.63–2.35) | ||||
Outer regional and | 0.83 (0.43–1.6) | ||||
Remote | |||||
DIC | 6,923 | 6,206 | 6,247 |
. | . | Model B . | . | Model C . | Model D . |
---|---|---|---|---|---|
. | Model A . | Unadjusted OR . | . | Adjusted OR . | Adjusted OR . |
. | (No covariate)a . | (95% Crl)a . | DIC . | (95% Crl)a . | (95% Crl)a . |
Individual level | |||||
Lung cancer types | 6,864 | ||||
SCLC | Reference | Reference | Reference | ||
NSCLC | 0.45 (0.37–0.55)b | 0.46(0.38–0.56)b | 0.52(0.43–0.63)b | ||
Stage | 6,668 | ||||
Stage I | Reference | Reference | Reference | ||
Stage II | 0.22 (0.16–0.29)b | 0.21 (0.15–0.28)b | 0.21 (0.15–0.28)b | ||
Stage III | 0.19 (0.15–0.24)b | 0.19 (0.14–0.24)b | 0.18 (0.14–0.24)b | ||
Stage IV | 0.31 (0.24–0.39)b | 0.36 (0.28–0.46)b | 0.34 (0.27–0.44)b | ||
Lung cancer types and stage (interaction) | 6,683 | ||||
SCLC | |||||
Stage IV | Reference | ||||
Stage I–IIIc | 0.45 (0.28–0.71)b,c | ||||
NSCLC | |||||
Stage IV | Reference | ||||
Stage IIId | 0.67 (0.58–0.78)b,d | ||||
Stage I–IId | 2.01 (1.7–2.39)b,d | ||||
Comorbidities | 0.97 (0.87–1.08) | 6,916 | 0.73 (0.64–0.83)b | ||
Smoking | 6,900 | ||||
Never | Reference | ||||
Ever | 0.74 (0.54–1) | ||||
Not stated | 1 (0.1–9.96) | ||||
ECOG PS | 6,564 | ||||
Excellent 0e | Reference | Reference | Reference | ||
Good 1f | 0.51 (0.43–0.61)b | 0.49 (0.41–0.59)b | 0.51 (0.42–0.61)b | ||
Moderate 2g | 0.2 (0.16–0.25)b | 0.18 (0.14–0.23)b | 0.19 (0.15–0.24)b | ||
Poor 3+4h | 0.08 (0.06–0.11)b | 0.07 (0.05–0.09)b | 0.07 (0.05–0.1)b | ||
Not stated | 0.43 (0.36–0.51)b | 0.36 (0.3–0.43)b | 0.37 (0.31–0.45)b | ||
Hospital insurance status | 6,863 | ||||
Private hospital | Reference | Reference | Reference | ||
Public hospital | 0.63 (0.55–0.73)b | 0.68 (0.57–0.83)b | 0.7 (0.57–0.84)b | ||
Hospital volume | 6,899 | ||||
1st quartile | Reference | ||||
2nd quartile | 0.88 (0.74–1.04) | ||||
3rd quartile | 0.82 (0.69–0.98)b | ||||
MDMs | 1.01 (0.9–1.13) | 6,918 | |||
Area level | |||||
Socioeconomic disadvantage | 6,906 | ||||
1 (most disadvantaged) | Reference | Reference | Reference | ||
2 | 1.09 (0.84–1.41) | 1.44 (1.09–1.95)b | 1.43 (1.09–1.93)b | ||
3 | 1.28 (0.98–1.7) | 1.51 (1.14–2.03)b | 1.48 (1.12–2.01)b | ||
4th quartile (least disadvantaged) | 1.24 (0.93–1.7) | 1.64 (1.19–2.31)b | 1.61 (1.14–2.37)b | ||
Remoteness | 6,840 | ||||
Major cities | Reference | ||||
Major city and inner regional | 1.09 (0.84–1.4) | ||||
Inner regional | 1.36 (0.95–1.98) | ||||
Inner regional and outer regional | 1.02 (0.67–1.57) | ||||
Outer regional | 1.2 (0.63–2.35) | ||||
Outer regional and | 0.83 (0.43–1.6) | ||||
Remote | |||||
DIC | 6,923 | 6,206 | 6,247 |
aStandardized for age–sex strata.
bStatistically significant (P < 0.05).
cReference category (SCLC stage IV).
dReference category (NSCLC stage IV).
eFully active (able to carry out all normal activity without restriction).
fRestricted in physically strenuous activity (ambulatory and able to carry out light work).
gAmbulatory and capable of all self-care (unable to carry out any work activities).
hCapable of only limited self-care (confined to bed or chair more than 50% of waking hours)/completely disabled (not able to self-care, totally confined to bed or chair).
With the inclusion of individual- and area-level risk factors, the likelihood of receiving GCT decreased significantly with poor ECOG PS (≥1), advanced clinical stages, non-small cell type, public hospital insurance, and comorbidities (models C and D). Patients resident in least disadvantaged areas were significantly more likely to receive GCT. We added covariates that considered an interaction between lung cancer types and clinical stages at diagnosis. Stage I–III SCLC (OR, 0.34; 95% Crl, 0.21–0.56) had a significantly lower likelihood of GCT compared with stage IV SCLC. Compared with stage IV NSCLC, stage III NSCLC had significantly lower GCT likelihood (OR, 0.57; 95% Crl, 0.49–0.66) and stage I–II NSCLC had a significantly higher likelihood (OR, 1.67; 95% Crl, 1.4–1.99). But, DIC was found to be lowest in the adjusted model C with spatial and temporal random effects and without interaction between lung cancer types and clinical stages.
Figures 1 and 2 show the age- and sex-standardized RR of GCT across LGAs. Larger RR indicates a higher risk of GCT. The maps display a substantial variation of RR across LGAs over the years. Geographic variability was higher in later years, and an increase in risk was observed over the years. Model estimates showed the rate of increase over the years was significant, but small in magnitude (4%). Regional areas outside of the metropolitan areas were likely to be more disadvantaged than outer regional/remote areas (Supplementary Fig. S1).
Discussion
Main findings
The overall level of adherence to GCT among SCLC and NSCLC was around 60.36%. It varied across geographic areas over the years. This variation was mainly associated with poor ECOG PS (≥1), followed by advanced clinical stages, NSCLC type, public hospital insurance status, area-level deprivation, and comorbidities.
Interpretation of results and comparison with other studies
On the basis of DIC and magnitude of OR, ECOG PS was the major contributor to GCT. Worse PS was associated with a lower likelihood of GCT, particularly for patients with advanced stage. Therefore, these findings highlight the importance of patient assessment by physicians before treatment because clinical guidelines are supposed to be followed in medically fit patients and might not be applicable in medically unfit patients. Previous studies adjusted for comorbidities, histology, and stage as fitness for treatment, but not for PS (7, 33, 34). Consistently with those studies, disease severity and comorbidities burden could partly explain the lower chances of receiving GCT.
Other studies in the United States reported that GCT was received by around 44.7%–76% of patients (7, 33, 34). Receipt of GCT was observed to vary by lung cancer type and stage, as these determine the selection of appropriate treatment. Consistent with this study (7), receipt of GCT was found to be lower among NSCLC type and advanced clinical stages. The rates of GCT were found to decrease with an advancing stage among NSCLC cases and increase with an advancing stage among SCLC cases.
SCLC is fast-growing cancer which spreads quickly, with very poor outcomes in untreated patients (35), which may lead to a higher likelihood of receiving more comprehensive treatment (7). In addition, limited (stage I–III) SCLC subjects were 66% less likely to receive GCT than patients with extensive (stage IV) SCLC. A higher proportion of patients with comorbidities, smokers, and public hospital insurance, and those residing in most disadvantaged areas among stage I–III SCLC could also have explained the lower likelihood of GCT.
Among NSCLC cases, localized (stage I–II) NSCLC received higher GCT than stage III and IV NSCLC, and this may relate to more straightforward patient selection and increased expectation of curative outcome. But, rates of GCT were 43% lower for locally advanced (stage III) than advanced (stage IV) NSCLC. Patients with locally advanced NSCLC were more likely to be smokers, have public hospital insurance, and stay at most disadvantaged areas, and this could explain the lower GCT uptake.
Most localized NSCLC cases received surgery for GCT, whereas SBRT and other treatment types were delivered much less commonly. Most cases in the potentially operable subgroups, locally advanced NSCLC, did not receive surgery as GCT. Chemotherapy was most commonly received among patients with late-stage, advanced NSCLC and extensive SCLC.
In this study, 39.64% of cases received less-intensive-than-recommended treatment, and 81.96% of those were NSCLC stage III–IV. Despite the availability of multiple treatment options, many patients (14%) did not receive active treatment in this study. The majority of untreated patients had NSCLC histology or late-stage disease. As most patients not receiving GCT presented with advanced stage, quality-of-life (QOL) rather than survival could have become an important consideration. Although poor PS or late stage could have deterred active treatment, 43.81% and 13.51% of those not receiving GCT were of good PS (ECOG, 0 or 1) and NSCLC stage I–II. Other factors could have contributed to not receiving GCT or no treatment. Possible explanations include a lack of referral to an oncologist (36–38) or MDM (39), known to increase active treatment, ageist and nihilistic attitudes toward lung cancer treatment (40), and patient’s preference against chemotherapy or radiotherapy (37). A previous study reported that the patient’s preference against treatment was the most common reason for not seeking a specialist opinion (37). Chemotherapy for late-stage lung cancer has been reported to improve QOL and survival and control symptoms compared with best supportive care (41). Thus, a multi-disciplinary team approach involving oncologists, surgeons, and palliative care specialists in the care process with shared decision-making can incorporate clinical judgement and patient preferences. It could help to reduce the proportion of cases receiving no treatment or less-intensive-than-recommended treatment.
Longer waiting times for lung cancer diagnosis and treatment among patients managed in public hospitals (42) as well as more advanced stage (III–IV) patients admitted to public hospitals could explain the lower likelihood of receiving GCT among those with public hospital insurance status. Other studies found socioeconomic factors, such as lack of insurance, low median household income, and geographic region, were related to lower adherence to GCT and poor survival (7–9). A possible explanation for socioeconomic disparities in the uptake of GCT includes financial burden, out of pocket and indirect costs, and longitudinal, protracted nature of combination therapy (43). Individuals living in disadvantaged areas had a higher risk of lung cancer due to a higher prevalence of risk factors like smoking (44). They were often disadvantaged with access to healthcare services (44). The vast majority of patients with low SES, indigenous status, and those in remote areas was diagnosed late with advanced stage (44). Thus, even after controlling for prognostic factors, such as lung cancer types, stage, PS, comorbidities, and hospital insurance, there was still a contextual effect of area-level deprivation on low GCT uptake. In addition, improving access to comprehensive quality care across public hospitals could help address inequalities in the delivery of GCT to patients with public hospital insurance, which might further reduce disparities related to socioeconomic deprivation.
Strengths
Within the Bayesian spatial-temporal multilevel framework, this study assessed the joint, independent effects of individual- and area-level factors to explain variation in individual- and area-level outcomes through a single, parsimonious framework. Our model integrated the spatial composition and temporal variation within a multilevel model that maintains the nested structure of a highly complex dataset.
While accounting for within-areas correlations and between-areas spatial and nonspatial associations and linear and nonlinear trends, the integrated models also incorporate within-area variability of covariates. It can help increase precision, distinguish between individual and contextual effects, reduce ecological bias, and allow for individual-level inference (23, 24, 45). This modeling framework provides more unbiased estimates of individual-level effects. It reduces ecological bias due to model misspecification, and avoids identifiability problems in ecological study. It increases statistical power compared with the standard multilevel model using individual-level data only (23). Like other studies (24, 45), our findings indicate a lack of statistical power of the individual-level data alone for investigating the independent effect of the individual- and contextual-level covariates when fitting the standard multilevel model to the outcomes from individual-level data. The standard multilevel model results show a larger magnitude of the OR for some covariates, but the direction of association was consistent with Table 4 results (Supplementary Table S1).
In addition, these models can be applied to look at geographic inequalities in outcomes of other diseases in different settings. Particularly, the models consider the heterogeneity of disease-, patient-, and institution-related factors to explain the spatial-temporal variation of GCT. The measures of IRSD and remoteness indices from ABS are robust and well-recognized indicators at the area level.
Limitations
This study retrospectively analyzed data which limits the capacity for causal inference in our conclusions. Because individual-level SES was unavailable, we analyzed socioeconomic, remoteness factors of the areas in which the patients lived at the time of diagnosis. We used VLCR data which might limit generalization of findings to other Australian hospitals that did not contribute to the VLCR. However, distribution of age, gender, and stage of disease in VLCR was relatively similar to that of national cancer statistics. There could be other unaccounted factors like physician and patient beliefs and preferences, physician–patient communication, or other unmeasured factors which may affect our study results.
Implications for policy and practice, and further research
This study quantified the contribution of the patient-, hospital-, and area-level characteristics to the geographic and temporal variation at the LGA level where health planning decisions are made (20). Because providers influence treatment utilization besides patient’s decisions, this study suggests future interventions to target providers, as well as to help improve disparities in guideline concordance. These disparities had impacts on patients’ outcomes as the receipt of guideline-concordant lung cancer care significantly improved survival outcomes among NSCLC and SCLC (7–9, 34) and survival times were significantly shorter in the guideline-discordant group in this study (Supplementary Fig. S2).
We also applied the Bayesian spatial-temporal multilevel models on 2-year mortality outcomes of patients diagnosed from 2011 to 2017. GCT was significantly associated with 43% lower risk of 2-year mortality outcomes (OR, 0.57; 95% Crl, 0.5–0.66) after adjusting for clinical stage, lung cancer type, ECOG status, smoking, MDM, hospital insurance, and socioeconomic disadvantage. Thus, quality benchmarking of guideline adherence would be useful as an actionable target to reduce inequalities in the receipt of curative cancer treatment and survival outcomes.
Despite the presence of evidence-based clinical guidelines for lung cancer management from various organizations, there is still limited adoption of these guidelines in clinical practice. The resulting disparities in receipt of guideline-concordant lung cancer care highlight the disproportionate burden of lung cancer treatment among those clinical subgroups with public hospital insurance and in deprived areas. Thus, interventions aimed at reducing these observed disparities in lung cancer treatment would help in improving the overall quality of lung cancer care and long-term survival outcomes. Future research can identify the impact of physician’s treatment choices and patient’s treatment preferences on GCT.
Conclusion
The overall level of GCT utilization among SCLC and NSCLC was around 60.36% and varied across geographic areas over the years. This variation was mainly associated with poor ECOG PS, followed by advanced clinical stages, NSCLC types, public hospital insurance status, area-level deprivation, and comorbidities. This study highlights the need to address disparities in receipt of GCT among patients with lung cancer with poor PS, NSCLC, advanced clinical stage, stage I–III SCLC, stage III NSCLC, public hospital insurance, comorbidities, and living in socioeconomically disadvantaged areas. Two-year mortality significantly improved with GCT. Thus, interventions aimed at reducing these inequalities could help to improve lung cancer outcomes.
Disclosure of Potential Conflicts of Interest
No potential conflicts of interest were disclosed.
Disclaimer
This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.
Authors' Contributions
W. Wah: Conceptualization, data curation, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. R.G. Stirling: Validation, writing–review and editing. S. Ahern: Supervision, validation, writing–review and editing. A. Earnest: Supervision, validation, writing–review and editing.
Acknowledgments
The authors thank Margaret Brand and the Victoria Lung Cancer Registry for registry data collection, management, and retrieval.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
Note: Supplementary data for this article are available at Cancer Epidemiology, Biomarkers & Prevention Online (http://cebp.aacrjournals.org/).