If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
Lung Immune Therapy Evaluation (LITE) Risk, A novel prognostic model for patients with advanced non-small cell lung cancer treated with immune checkpoint blockade
Corresponding author: Dr. Vishal Navani Division of Medical Oncology Department of Oncology University of Calgary 1331 29th St. NW, Calgary, AB T2N 4N2, Telephone: +1 825 994 2463
Department of Medical Oncology, Tom Baker Cancer Centre, Calgary, Alberta, CanadaDepartment of Oncology, Cumming School of Medicine, University of Calgary, Calgary, Alberta, Canada
Department of Community Health Sciences, Cumming School of Medicine, University of Calgary, Calgary, Alberta, CanadaDepartment of Cancer Epidemiology and Prevention Research, Alberta Health Services, Calgary, AB, CanadaForzani & MacPhail Colon Cancer Screening Centre, University of Calgary, Calgary, Alberta, Canada
Department of Oncology, Cumming School of Medicine, University of Calgary, Calgary, Alberta, CanadaDepartment of Community Health Sciences, Cumming School of Medicine, University of Calgary, Calgary, Alberta, CanadaDepartment of Cancer Epidemiology and Prevention Research, Alberta Health Services, Calgary, AB, Canada
Department of Community Health Sciences, Cumming School of Medicine, University of Calgary, Calgary, Alberta, CanadaDepartment of Cancer Epidemiology and Prevention Research, Alberta Health Services, Calgary, AB, CanadaForzani & MacPhail Colon Cancer Screening Centre, University of Calgary, Calgary, Alberta, Canada
Department of Medical Oncology, Tom Baker Cancer Centre, Calgary, Alberta, CanadaDepartment of Oncology, Cumming School of Medicine, University of Calgary, Calgary, Alberta, Canada
Department of Medical Oncology, Tom Baker Cancer Centre, Calgary, Alberta, CanadaDepartment of Oncology, Cumming School of Medicine, University of Calgary, Calgary, Alberta, Canada
Department of Oncology, Cumming School of Medicine, University of Calgary, Calgary, Alberta, CanadaDepartment of Community Health Sciences, Cumming School of Medicine, University of Calgary, Calgary, Alberta, CanadaDepartment of Cancer Epidemiology and Prevention Research, Alberta Health Services, Calgary, AB, Canada
Department of Medical Oncology, Tom Baker Cancer Centre, Calgary, Alberta, CanadaDepartment of Oncology, Cumming School of Medicine, University of Calgary, Calgary, Alberta, Canada
Department of Medical Oncology, Tom Baker Cancer Centre, Calgary, Alberta, CanadaDepartment of Oncology, Cumming School of Medicine, University of Calgary, Calgary, Alberta, Canada
Question: Can readily available clinical data be used to inform a prognostic scoring model in patients with advanced non-small cell lung cancer (NSCLC) treated with immune checkpoint inhibitors (ICI)?
•
Findings: In this cohort study of 495 patients with NSCLC, a prognostic scoring system was derived and externally validated. Patients were parsed into 3 discrete risk groups. The final model consisted of the following baseline characteristics: Eastern Oncology Cooperative Group (ECOG) performance status, derived neutrophil to lymphocyte ratio (dNLR) and lactate dehydrogenase (LDH).
•
Meaning: A simple prognostic scoring tool utilizing accessible clinical data can discriminate survival outcomes in patients with advanced NSCLC treated with single-agent ICI.
Abstract
Background
Immune checkpoint inhibitors (ICI) have revolutionized non-small cell lung cancer (NSCLC). We aimed to identify baseline characteristics, that are prognostic factors for overall survival (OS) in patients with NSCLC treated with ICI monotherapy, in order to derive the Lung Immune Therapy Evaluation (LITE) risk, a prognostic model.
Methods
Multi-centre observational cohort study of patients with advanced NSCLC that received ≥1 dose of ICI monotherapy. The training set (n=342) consisted of patients with NSCLC who received 1st line ICI. The test set (n=153) used for external validation was a discrete cohort of patients who received 2nd line ICI. 20 candidate prognostic factors were examined. Penalized Cox regression was used for variable selection. Multiple imputation was used to address missingness.
Results
Three baseline characteristics populated the final model: ECOG (0, 1 or ≥2), lactate dehydrogenase>upper limit of normal, and derived neutrophil to lymphocyte ratio ≥3. Patients were parsed into 3 risk groups; favorable (n=146, risk score 0-1), intermediate (n=101, risk score 2) and poor (n=95, risk score ≥3). The c-statistic of the training cohort was 0.702 and 0.694 after bootstrapping. The test cohort c-statistic was 0.664.
The median OS for favorable, intermediate and poor LITE risk were; 28.3 months, 9.1 months and 2.1 months respectively. Improving LITE risk group was associated with improved OS, intermediate vs favorable HR 2.08 (95%CI 1.46–2.97, p<0.001); poor vs favorable HR 5.21 (95%CI 3.69–7.34, p<0.001).
Conclusions
A simple prognostic model, utilizing accessible clinical data, can discriminate survival outcomes in patients with advanced NSCLC.
In this cohort study of 495 patients with NSCLC, a prognostic scoring system was derived and externally validated. Patients were parsed into 3 discrete risk groups. The final model included baseline Eastern Oncology Cooperative Group performance status, derived neutrophil to lymphocyte ratio and lactate dehydrogenase. A simple prognostic scoring tool utilizing accessible clinical data can discriminate survival outcomes in patients with treated with single-agent ICI.
Introduction
Immune checkpoint inhibitors (ICIs) have drastically altered the treatment paradigm for non-small cell lung cancer (NSCLC) over the past decade
Five-Year Outcomes From the Randomized, Phase III Trials CheckMate 017 and 057: Nivolumab Versus Docetaxel in Previously Treated Non-Small-Cell Lung Cancer.
Outcomes of patients with solid tumour malignancies treated with first-line immuno-oncology agents who do not meet eligibility criteria for clinical trials.
Effectiveness and Safety of First-Line Pembrolizumab in Older Adults with PD-L1 Positive Non-Small Cell Lung Cancer: A Retrospective Cohort Study of the Alberta Immunotherapy Database.
. As such, robust prognostic scoring tools are needed to aid with risk stratification in order to engage in data-driven shared decision making with patients. Given this evidence gap, we sought to derive and validate a simple and clinically accessible prognostic model, termed the lung immune therapy evaluation (LITE)-risk, to stratify patients with advanced NSCLC treated with ICI monotherapy.
Methods
Study Design
The Alberta Immunotherapy Database (AID) is a multi-centre observational cohort study. Baseline demographic, clinical and tumour characteristics, alongside survival outcomes are identified using a standard template and collected retrospectively
Effectiveness and Safety of First-Line Pembrolizumab in Older Adults with PD-L1 Positive Non-Small Cell Lung Cancer: A Retrospective Cohort Study of the Alberta Immunotherapy Database.
. This database captures outcomes from patients from the time of ICI monotherapy initiation. The training cohort consisted of treatment naïve patients with histologically confirmed advanced NSCLC that received first line (1L) pembrolizumab monotherapy, primarily with PD-L1 tumour proportion score (TPS) ≥ 50%. The test cohort included patients treated with nivolumab or atezolizumab in the second line (2L). We sought to identify baseline characteristics at ICI commencement, independently associated with inferior OS in order to derive a prognostic risk model, initially using the training cohort and then validated on the test cohort. 20 candidate variables were examined based on known clinically accessible, baseline characteristics, with reported prognostic or predictive associations with OS from the contemporary literature
Effectiveness and Safety of First-Line Pembrolizumab in Older Adults with PD-L1 Positive Non-Small Cell Lung Cancer: A Retrospective Cohort Study of the Alberta Immunotherapy Database.
Association Between Smoking History and Overall Survival in Patients Receiving Pembrolizumab for First-Line Treatment of Advanced Non-Small Cell Lung Cancer.
Patients initiated treatment between January 2010 and December 2019, with data analysis undertaken in February 2022. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines were followed. This study was approved by the Health Research Ethics Board of Alberta – Cancer Committee (HREBA.CC-19-0380).
Outcomes of Interest
The primary endpoint was overall survival (OS), defined as the time from commencement of ICI to death from any cause or date of last follow-up.
Statistical Analysis
Baseline laboratory markers were dichotomized based on their relationship to normal ranges and continuous variables based on conventional cut points. All variables were binary in nature except Eastern Cooperative Oncology Group (ECOG) performance status, where patients were stratified as ECOG = 0, ECOG = 1, or ECOG ≥ 2. For this study, we used least absolute shrinkage and selection operator (LASSO) Cox regression for variable selection. LASSO is a statistical method that performs parsimonious variable selection and helps to improve the generalizability of the model to external datasets by reducing the degree of overfitting
. To deal with missing data, we used multiple imputations by chained equations (MICE). This approach is commonly used and is less biased compared to a complete case analysis that excludes all patients with missing data
. LASSO Cox regression was fit on the original dataset through 10-fold cross-validation to identify the shrinkage parameter λs that yielded a 3-, 4-, or 5-variable model. Cross-fold validation is a resampling method that uses different portions of the dataset to for testing and training in order to predict how a model will perform on an independent dataset that was not used to derive it. In this way an attempt is made to avoid overfitting or selection bias.
The shrinkage parameter controls the degree to which the model coefficients are shrunk towards zero and number of covariates selected. Model performance was assessed using Harrell's C-statistic. The C-statistic gives the probability a randomly selected patient who experienced an event (death) had a higher risk score than a patient who had not experienced the event. A value of 0.5 means that the model is no better than predicting the outcome than random chance, where values of 0.7 or higher indicate a good model.
The prognostic risk model generated was built with independent prognostic factors using the training cohort and then validated using the test cohort. It has been established that prognostic models such as this perform better on data on that informed model construction compared to new data
. Because of this, internal and external validation are required. Internal validation was carried out internally using 500 iterations of bootstrapped samples in the training cohort. Bootstrapping is a statistical procedure that resamples a single dataset with replacement in order to create many simulated samples
. The same variable selection process through LASSO Cox regression was carried out in each bootstrap sample to develop a prognostic model. C-statistics of this model on the bootstrap sample (bootstrap performance) and in the original sample (test performance) were calculated. Optimism, the difference between training error and test error, was estimated as the mean difference between the bootstrapped performance and the test performance after 500 iterations. External validation was conducted using an independent test set comprised of individuals who initiated 2L systemic therapy with an ICI following a 1L therapy other than pembrolizumab. No patients in the training set were included in the test set. OS curves were estimated using the Kaplan-Meier method.
All statistical analyses were performed in R v.4.0.2 (R, Vienna, Austria). All statistical tests were 2 sided with a significance level of ≤0.05.
Results
342 systemic therapy naïve patients receiving pembrolizumab were examined in the training cohort. The external validation cohort (n=153) consisted of patients treated in the 2L with nivolumab (95.4%) or atezolizumab (4.6%) monotherapy As per Table 1, notable significant differences between the test cohort and training cohorts included a lower proportion of patients with adenocarcinoma (62.1% vs 72.8% p=0.05) and PD-L1 TPS ≥ 50% (2.6% vs 95% p<0.0001). More patients in the validation cohort had metastatic disease at ICI initiation (92.2% vs 84% p=0.04) and a lactate dehydrogenase (LDH) > upper limit of normal (ULN) (41% vs 29.3% p=0.029).
Table 1Baseline Characteristics of Training & Test Cohorts
The median follow-up time in the training cohort was 21 months (95% CI 19.6 – 22.9).
One clinical (ECOG) and two laboratory (derived neutrophil to lymphocyte ratio [dNLR] ≥ 3 and lactate dehydrogenase [LDH] ≥ ULN) variables were included in the final multivariable model, based on LASSO Cox regression and independent adverse association with OS, Table 2. The 20 candidate baseline candidate clinical and laboratory variables examined are outlined in Table 3.
Table 2Multivariable Cox regression analyses of final clinical variables derived from the training cohort (n=342)
Figure 1A outlines the negative prognostic relationship seen with an increasing risk score in the training cohort.
Figure 1(A). Kaplan Meier Overall Survival curve for patients in the training cohort, parsed by individual risk score. (B). Kaplan Meier Overall Survival curve for patients in the training cohort, parsed by a 3 risk group model (favorable, intermediate and poor risk). (C). Kaplan Meier Overall Survival curve for patients in the test cohort, parsed by a 3 risk group model (favorable, intermediate and poor risk)
Model validation indicated that 3-, 4-, and 5-factor models had similar optimism-corrected performance, while the 3-factor model displayed the best performance in the training cohort (Tables 5&6). Therefore, we developed a risk score-based prognostic model based on these 3 baseline variables. The presence of each variable received a score of 1, except for ECOG, in which patients with ECOG ≥ 2 received a score of 2. After comparing several prognostic groupings, a 3 group approach, termed LITE-risk gave the optimum combination of a high C-statistic, 0.702 (SE 0.017) and meaningful discrimination of patient outcomes (Figure 1B,Table 4,Table 7) with LITE favorable risk (n=146, risk score 0-1) median OS (mOS) 28.3 months (95% CI 20.8 – NE), LITE intermediate risk (n=101, risk score of 2) mOS 9.1 months (95% CI 7.1 – 15.3) and LITE poor risk (n=95, risk score ≥ 3) mOS 2.1 months (95% CI 1.7 – 3.7).
Table 4Overall survival outcomes and performance of the final prognostic model in both the training and test cohorts
Compared to the favorable risk group, both intermediate risk (HR 2.08; 95% CI 1.46 – 2.97) and poor risk (HR 5.21; 95% CI 3.69 – 7.34) had a significantly increased risk of death (p < 0.001 for both comparisons).
Model Validation
We undertook internal validation on the training set with 500 random bootstrapped resampled datasets to generate an optimism corrected C-statistic of 0.694 (95% CI 0.653 – 0.735), implying high discriminatory ability of the model.
The 3-factor model was then externally validated in the 2L test set, Table 4 and Figure 1C outline that the model consistently discriminated survival outcomes of patients. Compared to the favorable risk group, both intermediate risk (HR 1.82; 95% CI 1.16 – 2.86) and poor risk patients (HR 3.40; 95% CI 2.18 – 5.29) had a significantly increased risk of death (p < 0.001 for both comparisons), C-statistic 0.664 (SE 0.024).
Discussion
The findings of this study suggest that a simple and clinically accessible prognostic model can parse patients with advanced NSCLC treated with single-agent ICI into three discrete LITE-risk groups with clinically impactful differing OS, LITE intermediate vs favorable risk (HR 2.08; 95% CI 1.46 – 2.97) and poor vs favorable risk (HR 5.21; 95% CI 3.69 – 7.34), c-statistic 0.702. Consistent model performance was seen in the internally validated training and externally validated test cohorts.
The final 3-variable model included baseline patient ECOG performance status, dNLR and LDH; all of which have previously been described as being independent prognosticators of poor OS in advanced NSCLC
Effectiveness and Safety of First-Line Pembrolizumab in Older Adults with PD-L1 Positive Non-Small Cell Lung Cancer: A Retrospective Cohort Study of the Alberta Immunotherapy Database.
Differential prognostic effect of systemic inflammation in patients with non-small cell lung cancer treated with immunotherapy or chemotherapy: A post hoc analysis of the phase 3 OAK trial.
Modulation of peripheral blood immune cells by early use of steroids and its association with clinical outcomes in patients with metastatic non-small cell lung cancer treated with immune checkpoint inhibitors.
, our work adds value by providing external validation in a patient cohort treated in the second line setting. Prognostic factors should be intrinsic and not affected by systemic treatment. The clinically meaningful performance of this model in the 1L and 2L settings, which have established differences in baseline characteristics that impact ICI activity, such as level of PD-L1 expression and histology
give us confidence in the clinical relevance of LITE-risk groups. Our optimism adjusted c-statistic values in the training and test sets are also comparable to other prognostic models in this context
. It is important to note that our statistical approach increased internal validity via bootstrap analysis, which was not undertaken in other similar studies
. LITE-risk groups are able to provide clinicians new benchmarks for real world OS.
Previous prognostic models in this context such as LIPS-39 utilized the same datasets for training and validation, and this single split of datasets have been established to give erroneous estimation of model performance
. Our use of bootstrapping for internal validation, LASSO Cox regression to reduce overfitting and reporting of an optimism-corrected C-statistic improves the estimation of the generalization performance of LITE-risk when compared to LIPS-3
On Splitting Training and Validation Set: A Comparative Study of Cross-Validation, Bootstrap and Systematic Sampling for Estimating the Generalization Performance of Supervised Learning.
. We were not able to formally compare our model with LIPS-3 to determine statistical robustness due to the lack of data regarding pre-treatment steroids, a LIPS-3 predictor, in our dataset.
The inferior survival outcomes seen in the LITE-poor risk groups in both the training and validation cohorts, mOS of 2.1 and 3.1 months, respectively are striking. These data are helpful to allow informed conversations about expectations of therapy, and provides further justification for the early involvement of palliative care
Association of Early Palliative Care Use With Survival and Place of Death Among Patients With Advanced Lung Cancer Receiving Care in the Veterans Health Administration.
. It is unclear whether these patients would benefit from treatment intensification with chemo-immunotherapy.
Limitations
The retrospective nature of the work, and accrual of all patients from one province may limit the generalizability of our results. The lack of a control group, receiving combined chemo-immunotherapy or chemotherapy alone prevents us from assessing any predictive capabilities of LITE-risk. Our focus on overall survival as the primary outcome, and data capture only from time of ICI initiation, means that delineating the benefit of any subsequent lines of chemotherapy (for the 1L training cohort) from the survival gain provided by ICI is difficult.
Due to the limited patient population studied, it is not clear if our prognostic model applies in patients treated with other systemic therapies alone such cytotoxic chemotherapy or in other, earlier disease states amenable to curative intent ICI
after definitive chemoradiation. Our use of a 2L cohort as an external validation population may be controversial, due to the known differences in baseline characteristics and survival outcomes compared to an ICI naïve setting and thus prospective validation from larger, international datasets are required. PD-L1 expression is an established predictive biomarker in the 1L
Atezolizumab versus docetaxel in patients with previously treated non-small-cell lung cancer (OAK): a phase 3, open-label, multicentre randomised controlled trial.
contexts. However, in our dataset the rate of PD-L1 expression “missingness” in the test cohort (2L) was 42.4% (65/153), as opposed to 0% in the training cohort (1L), as it is not a public reimbursement payor requirement prior to receipt of 2L therapy, as opposed to 1L therapy. Statistical approaches would be unreliable in this context as the data are not missing at random, making MICE an inappropriate tool, therefore we elected not to evaluate this as a candidate variable. The assessment of LITE-risk in association with PD-L1 expression should be evaluated prospectively to outline the prognostic/and or predictive value of this model.
A prognostic variable used, LDH was missing data in 34% of patients. However, when examining the impact of this missing data on variable selection, it is likely that LDH is missing at random, and so MICE is an appropriate tool
, however we intentionally focused on a LITE-risk score that was easy to derive, utilising commonly used a priori cut-offs. This provides parsimony and the ease of use should not be understated were this to be incorporated into practice. Statistical information and power is commonly lost with categorisation in this way, but is common in medicine.
Conclusions
This study describes the derivation and external validation of a simple prognostic LITE-risk model utilizing readily accessible clinical data points for patients with advanced NSCLC treated with ICI monotherapy.
Credit Author Statements
Dr. Vishal Navani had full access to all the data in the study and takes responsibility for the integrity of the data and accuracy of the data analysis. All information and material in the manuscript are original.
Dr Heng reported receiving grants from Pfizer, Novartis, Bristol-Myers Squibb, Ipsen, Merck, and Eisai outside the submitted work.
Dr Navani reports personal consulting fees from Kwoya Kirin, Novotech PTY, Pfizer, Astra Zeneca, EMD Serono and IPSOS
The other authors report no disclosures
Acknowledgements
A provinical statewide Alberta ethics (Health Research Ethics Board of Alberta) has provided overseeing ethics approval for this study (HREBA.CC-19-0380).
Data analysis and statistical code utilised can be made available on reasonable request
Five-Year Outcomes From the Randomized, Phase III Trials CheckMate 017 and 057: Nivolumab Versus Docetaxel in Previously Treated Non-Small-Cell Lung Cancer.
Outcomes of patients with solid tumour malignancies treated with first-line immuno-oncology agents who do not meet eligibility criteria for clinical trials.
Effectiveness and Safety of First-Line Pembrolizumab in Older Adults with PD-L1 Positive Non-Small Cell Lung Cancer: A Retrospective Cohort Study of the Alberta Immunotherapy Database.
Association Between Smoking History and Overall Survival in Patients Receiving Pembrolizumab for First-Line Treatment of Advanced Non-Small Cell Lung Cancer.
Differential prognostic effect of systemic inflammation in patients with non-small cell lung cancer treated with immunotherapy or chemotherapy: A post hoc analysis of the phase 3 OAK trial.
Modulation of peripheral blood immune cells by early use of steroids and its association with clinical outcomes in patients with metastatic non-small cell lung cancer treated with immune checkpoint inhibitors.
On Splitting Training and Validation Set: A Comparative Study of Cross-Validation, Bootstrap and Systematic Sampling for Estimating the Generalization Performance of Supervised Learning.
Association of Early Palliative Care Use With Survival and Place of Death Among Patients With Advanced Lung Cancer Receiving Care in the Veterans Health Administration.
Atezolizumab versus docetaxel in patients with previously treated non-small-cell lung cancer (OAK): a phase 3, open-label, multicentre randomised controlled trial.
00004-9&id=gr1.jpg){kind=link}