Measuring toxicity following systemic therapy for lung cancer in Aotearoa New Zealand using a cancer-specific framework and national collections data

Toxicities are common among those receiving systemic therapy as part of cancer treatment.^1,2 Toxicities are varied and can affect any body system, including fatigue, nausea, vomiting, rashes, diarrhoea, constipation, mucositis and venous thromboembolism, and myelosuppression including neutropenia, thrombocytopenia and anaemia.¹ While most adverse events are managed in the community, more serious cases will require acute hospital admission or even lead to death.^3,4 Best practice is to grade toxicity severity using the Common Terminology Criteria for Adverse Events (CTAE), with more serious toxicities graded 3–5.

Toxicities associated with systemic therapy affect a patient’s quality of life and physical health and may affect an individual’s ability to complete treatment.^5,6 Several studies have shown that patients experiencing severe adverse effects have a higher risk of treatment interruption, discontinuation and mortality.^7–9 To help guide treatment decisions, it is important that both the clinician and patient understand toxicity associated with cancer therapies. Better understanding of side effects and ability to predict severe toxicity requiring hospital-level treatment is also vital in building an accurate picture of the resources required in the management of cancers.^2,10

Despite this, there is little population-level evidence on the rates of toxicity following systemic therapy for those diagnosed with cancer. To date, most of the data available on toxicity of systemic therapy are derived from clinical trials;^11,12 however, these can be limited in their applicability to real-world clinical practice.² The tendency for trials to exclude older people or those with comorbidity—both groups at higher risk of toxicity—means that toxicity in clinical trials may not represent the expected profile in the general population.¹³

Lung cancer is the most frequently diagnosed cancer and the leading cause of cancer deaths worldwide.¹⁴ The current management of lung cancer is rapidly changing, but includes a combination of surgery, radiation therapy and systemic therapies such as chemotherapy, immunotherapy and targeted therapy.^15,16 Treatment-modality decisions are based on a number of factors including type of lung cancer (i.e., small cell vs non-small cell), stage at diagnosis, comorbidities and fitness for therapy, including performance status. As early disease is often asymptomatic, most cases of lung cancer are advanced stage at diagnosis.¹⁷ Patients with lung cancer also tend to have high levels of comorbidity and poor baseline health at diagnosis.¹⁸ Due to these factors, patients with lung cancer are often not suitable for surgical management, and systemic and radiation therapies are a common form of treatment for disease control and palliation of symptoms.¹⁹

There is limited information about the types and rates of real-life toxicity among those treated for lung cancer outside of randomised controlled trials (RCTs). The under-representation of elderly and more comorbid groups in RCTs also means we have limited information on toxicity in these groups, despite knowing the disproportionate rates of comorbidity in patients with lung cancer.^20,21 One of the primary reasons we have limited information on toxicity among those with lung cancer is the lack of available national-level toxicity data. To that end, researchers in the universal healthcare context of the United Kingdom have developed a cancer-specific toxicity coding framework,²² using similar national-level administrative data to that collected in New Zealand. This framework may provide an opportunity to measure and report on toxicities experienced by those treated for lung cancer in New Zealand.

As such, the objective of this study was to use national-level hospitalisation data to describe the number and types of severe acute toxicities experienced by patients receiving systemic therapy for lung cancer in New Zealand, using the cancer-specific toxicity coding framework developed by Boyle et al.²² Our aim was to use this framework to gain an understanding of different types of severe toxicity within New Zealanders receiving systemic therapy for lung cancer.

Methods

Participants and data sources

We identified all lung cancer registrations (International Statistical Classification of Diseases and Related Health Problems, 10th Revision, Australian Modification [ICD-10-AM] codes: C33–C34) from 2012 to 2019 as listed on the New Zealand Cancer Registry (NZCR; N=18,081). This study period is prior to the introduction of publicly funded immunotherapy for lung cancer, so will primarily capture toxicities following other forms of systemic therapy. The NZCR is a mandatory record of all malignant cancers diagnosed in New Zealand except basal and squamous cell skin tumours.²³ This NZCR cohort were then linked via encrypted National Health Index number (NHI) to national health data, including public and private hospital inpatient records (National Minimum Dataset, NMDS); the national Pharmaceutical Collection (PHARMS); the Radiation Oncology Collection (ROC), which is a national collection of radiation therapy attendances; and the Mortality Collection (death registrations). We also linked NHI data to primary health organisation (PHO, primary care enrolments) data and National Non-Admitted Patient Collection (NNPAC, outpatient and emergency department attendances) data; these together with Mortality and NMDS data were used to supplement demographic information where NZCR records were incomplete. For example, domicile code was derived from the NZCR, except for 27 registrations with non-specific domicile codes (either “overseas” or a district health board area) where domicile information was sourced from other national collections datasets.

Once this cohort was linked to health data, we identified those who accessed systemic therapy with the PHARMS dataset using relevant chemical codes identified by our clinical advisors, with these codes listed elsewhere.²⁴ This list of codes was used to scan the PHARMS data for instances of systemic therapy receipt in the 90 days prior to cancer diagnosis (to allow for some therapy to have occurred prior to the diagnosis date that was recorded on the cancer registry) and up to 2 years post–cancer diagnosis (to allow for the commencement and completion of at least one course of systemic therapy). Those who did not access funded systemic therapy over this time period were excluded from further analysis; for this reason, immunotherapy was not included, since no immunotherapies were publicly funded over our study period.

Identifying severe acute toxicities

We used the coding framework developed by Boyle et al.²² to define severe acute toxicity following treatment with systemic therapy. This framework uses diagnostic codes with hospitalisation (NMDS) data to define the presence of a severe acute toxicity. Every ICD code in the Boyle framework was checked against the codes available in the ICD-10-AM (2nd edition) coding available for the current study to ensure correct mapping, with some minor modifications made following this review (Appendix Table 1).

Categories of toxicity were directly derived from the framework and included gastrointestinal and liver injury, infection, cardiovascular, metabolic/endocrine, constitutional (e.g., fatigue), renal, haematological, pain, respiratory, neurological, line complications, psychological, bleeding, dermatology/rheumatology, ophthalmic and drug reaction. Unlike Boyle et al.,²² we purposefully did not include death as a toxicity within the current study, as death within 1 year of treatment onset will be a common outcome for this cohort, and we could not, with the available data, discern whether any deaths were attributable to systemic therapy toxicity, disease progression or other causes. In line with the framework presented by Boyle et al.,²² some conditions that have common causes other than treatment with systemic therapy (e.g., anaemia, heart failure) were only counted as a treatment-related toxicity if they had not occurred in the 12-month period prior to the onset of systemic therapy. To take into account potential systematic differences in admission criteria between regions, we included diagnostic codes for all inpatient stays, regardless of duration. This deviates from the criteria used by Boyle et al.,²² where only overnight stays or longer were included.

To set the time period within which we would scan for evidence of toxicity-related admissions, we set the date of the first systemic therapy dispensed to each patient as their index “start date”. We then scanned forward for up to 1 year after this start date and set the “end date” as the date of the last systemic therapy dispensing plus 8 weeks, as per the Boyle et al. framework,²² to allow for the occurrence of toxicities after delivery of the last systemic therapy. The resulting start and end dates defined the “toxicity period” for a given patient (i.e., it was possible for patients to have a larger or smaller toxicity period depending on the time between their first and last systemic therapy dates). If a given toxicity occurred during the toxicity period, a binary flag (yes/no) was set to indicate whether a patient experienced that toxicity.

Demographic variables

Age at lung cancer diagnosis was derived from the NZCR and categorised as <50, 50–64, 65–74 and 75+ years. No cases were missing data on age. Prioritised ethnicity was derived from the NZCR or supplementary sources (as outlined above) and categorised into Māori, Pacific, Asian, European or Other. No cases were missing data on ethnicity. Area-level socio-economic deprivation was defined using the New Zealand Index of Deprivation (NZDep)²⁵ and categorised into five quintile categories (NZDep 1 [least deprived], 2, 3, 4, 5 [most deprived]). Missing data prevented the attribution of NZDep quintile for 80 diagnoses (0.3% of cohort). Rurality was defined using the Geographic Classification for Health (GCH), based on domicile code,²⁶ and categorised as either urban or rural. Missing data prevented the attribution of rurality for 195 diagnoses (0.7% of cohort). Comorbidity at diagnosis was defined using the C3 cancer comorbidity index²⁷ and categorised as 0 (score ≤0), 1 (>0, ≤1), 2 (>1, ≤2) and 3 (>2). If no recorded comorbidity was found over the 5-year period prior to diagnosis, individuals were assigned a score of 0. The C3 index was used as a categorical variable for descriptive analysis and as a splined variable for marginal standardisation using restricted cubic splines with three knots placed at the 75th, 90th and 95th percentiles.²⁸ No participants were missing comorbidity status.

Tumour variables

Tumour type was defined using morphology data from the NZCR and classified as either small cell lung cancer (SCLC), non-small cell lung cancer (NSCLC), or “other/unspecified”.²⁹ Our analysis focusses on SCLC and NSCLC tumours. No participants were missing data on tumour type, noting the lack of specific detail within the “other/unspecified” group. Stage of disease was extracted from the NZCR using the SEER Summary Stage method (A to F),³⁰ which was further aggregated into Local (SEER Summary Stage B), Regional (C and D), Advanced (E) and Unstaged (F). For lung cancer, SEER Stage is categorised as follows: “Local” refers to a single tumour confined to one lung; “Regional” refers to extension to nearby structures or lymph nodes; “Advanced” refers to distant metastasis or involvement of distant lymph nodes.³⁰ SEER Stage was used due to the incomplete nature of TNM (tumour, node and metastasis) stage reporting on the NZCR (as a non-mandatory field). “Unstaged” was treated as a meaningful stage category, and as such, no participants were missing stage information.

Controlling for other treatment

In order to control for the fact that patients may have been undergoing treatment(s) other than systemic therapy, and therefore that certain toxicities may be related to these other treatments rather than systemic therapy, we collected data on receipt of surgery and receipt of radiation therapy within our cohort. Receipt of surgery was defined using public and private hospitalisation (NMDS) data, with relevant curative and palliative procedures (as identified by our clinical advisors²⁴) extracted using Australasian College of Health Informatics (ACHI) ICD-10-AM codes (2nd edition). We set a binary indicator (yes/no) if one of the included surgeries occurred during the toxicity “window” outlined above, with an additional 90-day look-back period added prior to onset of systemic therapy, to allow for those cases where surgery was completed prior to onset of (adjuvant) systemic therapy. Receipt of radiation therapy was derived by scanning ROC data for receipt of radiation therapy and was set as a binary indicator (yes/no) if one of the radiation therapies occurred during the same “window” as for surgery. These two variables were used to define a further treatment combination variable, which was subsequently used to adjust within our marginal standardisation modelling (see Statistical analysis section below) for whether a given patient received systemic therapy only, systemic therapy plus surgery, systemic therapy plus radiation therapy, or a combination of all three.

Statistical analysis

Descriptive analysis

We completed crude, age-standardised and marginally standardised descriptive analyses to describe the number and percentages of cases within the cohort who experienced each of the toxicities across the total cohort, total cohort stratified by tumour type, and total cohort stratified by ethnicity. For percentages, the numerator was the number of cases who experienced the toxicity within the toxicity period, while the denominator was the total number of diagnosed cases within the cohort (or within a sub-group, e.g., tumour type or ethnicity). We determined crude numbers of cases and percentages using standard descriptive analysis. Age-standardised percentages were calculated through direct age standardisation using the PROC STDRATE procedure in SAS version 9.4 (SAS Institute Inc., North Carolina, United States of America [USA]), with the total Māori lung cancer population as the standard population. We used a Māori standard to normalise the age structure of the Māori population within our study.³¹ Within our results presentation, age-standardised percentages are labelled as “age adjusted %”.

Marginal standardisation

To control for potential confounding by factors such as comorbidity, stage and other treatment on the occurrence of a given toxicity, we used marginal standardisation to produce percentages adjusted for such covariates. These represent (based on re-expression from a logistic regression model) the expected toxicity profile from each of the compared groups if those groups had the same profile on the confounder variables. This marginal standardisation was calculated from logistic regression models of the outcomes (i.e., occurrence of a given toxicity) using the margins macro in SAS, with the total Māori lung cancer covariate profile as the reference. Variables included in the model were age, gender, ethnicity, deprivation, rurality, comorbidity, stage, tumour type and treatment combination. Tumour type and ethnicity were removed from their respective models when stratifying by these variables within our analysis. The rcspline macro was used to include the C3 comorbidity score as a splined variable within the marginal standardisation.³² Individuals missing data on deprivation or rurality were excluded from the marginal standardisation models (195 cases; 0.7% of cohort). Data were complete for all other variables included in the models.

Within our results presentation, marginally adjusted percentages are labelled as “fully adjusted”, which means that the proportion is adjusted for all modelled variables. For example, when comparing toxicity proportions between tumour-type groups, the fully adjusted proportion is adjusted for potential differences between these groups in age, gender, ethnicity, deprivation, rurality, comorbidity, stage of disease, tumour type and the combination of treatment that they received.

Analysis was completed using SAS version 9.4 (SAS Institute Inc., North Carolina, USA) and Microsoft Excel (Microsoft Corporation, Washington, USA).

Results

Cohort characteristics

A total of 5,098 patients received systemic therapy over the study period and were included in our analysis, and Table 1 describes their characteristics. The majority (71%) of patients were diagnosed with NSCLC, and the remainder had SCLC (26%) or other/unspecified (3%) tumour types. A total of 99 (2%) had local disease at diagnosis; 1,115 (22%) regional disease; 2,508 (49%) advanced disease; and 1,376 unstaged disease (27%), according to the cancer registry. Of those who received systemic therapy, 4,449 (87%) received chemotherapy and 812 (16%) received targeted therapy, with some patients in this cohort receiving both. Patients received either systemic therapy alone (59%), a combination of systemic therapy and radiation (38%), systemic therapy and surgery (2%) or a combination of all three treatment modalities (1%).

There was a relatively even split of females (52%) and males (48%) in our cohort, with the majority of patients aged between 65–74 (40%) and 50–64 (39%), with fewer patients in the older (75+) age groups (14%) and younger (<50-year-old) age groups (7%). Patients of European ethnicity made up 63% of the cohort, followed by Māori (25%), Asian (7%), Pacific (5%) and other (1%) ethnicities. A third of the cohort (31%) lived in the most deprived deciles (NZDep 9–10). A majority of patients (75%) resided in urban centres, with the remaining 25% residing in rural areas. In terms of comorbidity, 55% of patients had a C3 score of ≤0, 16% 0–1, 15% 1–2, and 14% >2 (Table 1).

Toxicity following systemic therapy in the total cohort

Of the 5,098 patients with lung cancer who received systemic therapy over the study period, about four in five (78% crude, 82% fully adjusted) experienced some toxicity (Table 2). Of all 17 toxicity categories, infections were the most common, with approximately half (47% crude, 55% fully adjusted) of the cohort having experienced some type of infection. The most common type of infections were lung infections (32% crude, 40% fully adjusted). Gastrointestinal (37% crude/fully adjusted), cardiovascular (34% crude, 39% fully adjusted), haematological (29% crude, 32% fully adjusted) and constitutional (26% crude, 29% fully adjusted) toxicities were the next most common.

Constipation was the most common gastrointestinal toxicity, with a fifth (21% fully adjusted) of the total cohort having experienced constipation following systemic therapy. The most common cardiovascular toxicity was hypotension (13% crude, 17% fully adjusted), followed by arrhythmia (13% crude, 16% fully adjusted). One fifth of the cohort developed anaemia (20% crude/fully adjusted) during the toxicity period, and around 15% (13% crude, 16% fully adjusted) developed neutropenia.

Toxicities by tumour type

The rates of toxicity were broadly similar for SCLC (82% crude, 84% fully adjusted) compared to NSCLC (76% crude, 81% fully adjusted; Table 3). Lung infection remained the most common toxicity across all tumour types, followed by electrolyte disturbances and fluid balance changes. Notable differences in toxicity patterns between tumour-type groups included the occurrence of neutropenia, which was almost twice as common among those with SCLC (crude 22%, fully adjusted 24%) compared to NSCLC (crude 10%, fully adjusted 12%). Sepsis was also more common among SCLC (crude 13%, fully adjusted 14%) compared to NSCLC (crude 8%, fully adjusted 9%), as was electrolyte disturbance (SCLC: crude 28%, fully adjusted 29%; NSCLC: crude 19%, fully adjusted 23%).

View Table 1–3.

Discussion

Application of the cancer-specific toxicity coding framework in New Zealand

This is the first time, to our knowledge, that hospitalisation data have been used to report on rates of toxicity following systemic therapy for lung cancer and for any type-F cancer in New Zealand. A key limitation is the reliance on accurate reporting and coding of diagnoses in hospitalisation data, since these data form the basis of the coding framework employed here. Related to this, it is important to note the distinction between toxicities that drive an admission to hospital and toxicities recorded during an admission; the current study reports the latter, and as such we cannot be certain that a toxicity was severe enough to warrant hospitalisation.

The commonality of some toxicities occasionally differed from what might have been expected. For example, the percentage of cases experiencing symptoms such as fatigue and cachexia were lower than expected based on clinical experience; this likely represents a limitation in our study design which only aimed to capture toxicity leading to hospital admission. By only evaluating the “tip of the iceberg” of serious toxicities (CTAE grade 3–5) leading to admission, we are limited by not knowing the rates and distribution of toxicity being managed in the community. In this way, our study may be biased toward capturing higher-grade toxicities. Further studies evaluating primary care data and self-reported toxicity would be useful to enhance our understanding of lower grade toxicities and would provide a broader view of adverse effects experienced in the community. Qualitative (interview) work to further describe the experience of toxicity among those with lung cancer in the real world would be of immense value and provide a richness to the data reported here.

Commonality of toxicity identified with the framework

Using the cancer-specific coding framework, we observed that toxicity is a common occurrence among those receiving systemic therapy for lung cancer, with 82% experiencing a hospital-diagnosed toxicity following onset of treatment. The most common toxicity experienced over this time was infection, with lung infections being the most common toxicity across tumour types and ethnic groups. This likely reflects a number of key underlying factors. Firstly, lung cancer has a high incidence of underlying airway and lung disease, often related to higher rates of smoking, which increases underlying risk of infection. Secondly, protective mechanisms against infection are also affected by the presence of cancer itself, as well as by immunocompromise secondary to systemic therapy, making serious infections leading to admission more likely. Lastly, it is possible that when a fever is without a known origin of infection, respiratory symptoms due to lung cancer could be classified as a lung infection.

We also noted a number of other toxicities that were common among our cohort which would likely have a major impact on quality of life. A fifth of those in our cohort were diagnosed with constipation (plausibly driven by systemic therapy and opioid medications), while a fifth were diagnosed with anaemia. Around one in every six patients were diagnosed with hypotension, one in six with arrythmia and one in 10 with sepsis—noting that each of these three conditions commonly coincide with each other.³³ The commonality of these toxicities reinforce the need for advanced clinical preparation for their occurrence following systemic therapy onset, including the communication of this likelihood with patients as is required in the process of informed consent. Furthermore, wrap-around care, clear cancer-service contact details, written reminders of potential toxicities and access to urgent medical attention are needed to either prevent or promptly recognise and manage these conditions as they occur.

We also note the commonality of gastrointestinal toxicities that may be less life-threatening than serious cardiac malfunction or sepsis but nonetheless carry immense distress for the patient, such as constipation (one in five patients). Such distress may be related to the commonality of opioid use among New Zealanders undergoing lung cancer treatment (43% of those with local disease, up to 74% of those with advanced disease³⁴), which is a useful reminder of the importance of laxative co-prescription and bowel care among those receiving systemic therapy for lung cancer.

Key findings by tumour type

While the overall burden of toxicity appeared similar between tumour types, those with SCLC were more likely to be diagnosed with neutropenia, sepsis and electrolyte disturbances. This difference in toxicity burden may be related to the aggressive nature of SCLC tumours compared to NSCLC tumours,³⁵ wherein the aggressiveness of the cancer itself is leading to heightened commonality of these conditions. This explanation is plausible since while we have adjusted for pre-treatment comorbidity within our fully adjusted analysis we were unable to adjust for health and performance status at the time of treatment (e.g., ECOG [Eastern Cooperative Oncology Group] status) using the available data. In addition, the heightened risk of neutropenia among those with SCLC may be related to differences in treatment type and resulting toxicity profile compared to NSCLC.¹⁹

Differences in the commonality of these toxicities between tumour types may also be driven by differences in the aggressiveness of treatment. A pooled analysis of three randomised trials found that survival outcomes among late-stage lung cancer patients with neutropenia are better than among those without neutropenia (approximately 43 weeks versus approximately 31 weeks), with the authors reasoning that the absence of neutropenia may be an indication of underdosing of systemic therapy.³⁶ In other words, the heightened commonality of neutropenia among those with SCLC in our cohort may be a clinically expected response to more aggressive systemic therapy dosing among this group, relative to those with NSCLC.

As noted in the introduction, there is limited information about the types and rates of real-life toxicity among those treated for lung cancer, outside of evidence from RCTs. It is difficult to compare levels of toxicity with other studies due to different source populations (New Zealand compared to international; national study compared to clinical trial/smaller centre studies; total lung cancer cohort compared to stage-specific and/or sub-type specific analysis) and different datasets used to define toxicity. With this in mind, the most common toxicities identified in the current study (e.g., anaemia, neutropenia, fluid imbalance, constipation) are also among the most commonly identified in other lung cancer contexts.^37–39

Strengths and limitations

Chemotherapy is often prescribed in combination with supportive medications such as antiemetics, steroids and opiate analgesics. We were not able to exclude these medications as the cause of toxicity rather than the systemic therapy itself. We also cannot rule out the possibility that toxicities reported here were the result of systemic therapies other than those captured within the current study (e.g., non–publicly funded immunotherapies). Relatedly, as noted in the Methods, the study “toxicity period” (which included treatments received up to the end of 2021) ended before public funding of immunotherapy medicines for lung cancer (mostly from 2023). As such, our findings largely reflect the toxicity profile within the pre-immunotherapy era of systemic therapy for lung cancer. While we have tried to capture all covariates that a) might act as confounders/mediators of the relationship between systemic therapy and toxicity and b) that are nationally available and complete, it remains plausible that other factors may have influenced the relationship between systemic therapies received in this study and resulting toxicity (e.g., clinical trial participation, as noted above with respect to non-funded systemic therapy access). Due to the overall poor prognosis of cases due to disease progression, it is plausible that some cases may have developed toxicity following systemic therapy (and had that toxicity recorded within our results) but died before that toxicity could be observed.

Finally, our study presents a broad picture of toxicity occurrence among patients with lung cancer, with limited stratification by tumour type and ethnicity. As such, our analysis remains a broad indication of the toxicity burden experienced by patients with lung cancer. It does not compare (for example) the occurrence of toxicity between those receiving one type of systemic treatment to another or experiencing one stage of disease within a given tumour type, since such analyses would likely require a nuanced approach beyond the broad focus of the current study. We are planning to undertake such work in the future, possibly with the addition of more granular treatment data as they become available.

Conclusions

We observed a significant toxicity burden using administrative health records and a cancer-specific toxicity coding framework: four in every five patients experienced some toxicity during the course of treatment, and around half of the cohort were diagnosed with an infection of some kind, most commonly lung infection. Constipation, hypotension, cardiac arrhythmia, anaemia and neutropenia were all highly common toxicities. Those with SCLC were more likely to be diagnosed with neutropenia, sepsis and electrolyte disturbances than those with NSCLC. This study used a cancer-specific toxicity coding framework and nationally available data to identify toxicities following systemic therapy among those with lung cancer. With the caveats identified and discussed throughout this paper, this system can be used to identify and better understand this burden in other cancer contexts.

View Appendix.