Model design

A Markov model was designed to simulate outcomes of a hypothetical cohort of advanced (stage IIIB/IV) NSCLC adult patients with untested EGFR-sensitizing mutation status (Fig 1). Markov model is an analytical framework in which the hypothetical patients proceed to mutually exclusive health states in the next model cycle according to transition probabilities, with costs and health outcomes aggregated over successive cycles. Four treatment strategies were evaluated in the present model: Empirical first-line chemotherapy with cisplatin-pemetrexed (empirical chemotherapy group) for all patients, and EGFR mutation-guided use of afatinib, erlotinib, and gefitinib. The model time horizon was 10 years (with monthly cycle) to allow adequate time for estimation of the lifelong outcome measures, including direct medical cost, progression-free survival, life-years and QALYs gained by each treatment strategy.

Fig 1

Simplified Markov model of empirical chemotherapy, EGFR mutation-guided afatinib, erlotinib, and gefitinib for treatment of advanced NSCLC.

EGFR: epidermal growth factor receptor; NSCLC: non-small-cell lung cancer.

Major health states in the Markov model included: Progression-free survival, disease progression and death. All patients entered the model at the health status of progression-free survival. The patients proceeded through health statuses in each monthly cycle according to probability inputs. In the empirical treatment group, the patients received cisplatin-pemetrexed for the up to six 21-day cycles [13]. In the EGFR-guided groups, patients who were tested positive for EGFR mutation were treated with gefitinib, erlotinib, or afatinib until disease progression occurred [5–7]. Patients with negative or undetermined EGFR-mutation testing results were treated with first-line chemotherapy. Palliative care, including palliative chemotherapy with pemetrexed and best supportive care, was provided to control symptoms for patients with disease progression [3].

Clinical inputs

All model inputs are listed in Table 1. Literature search on MEDLINE over the period 2000–2020 was performed using keywords including “advanced non-small-cell lung cancer”, “NSCLC”, “EGFR mutation”, “overall survival”, “progression-free survival”, “first-line treatment”, “first-line chemotherapy”, “gefitinib”, “erlotinib”, and “afatinib”. The selection criteria of clinical trials were: (1) reports in English language; (2) adult patients with stage IIIB/IV NSCLC; and (3) progression-free survival, overall survival or adverse event rates were reported. Preferred studies were meta-analyses or randomized controlled trials. When multiple randomized trials were available for the same model input, the weighted average was used as the base-case value and the high/low values formed as the upper/lower limits for sensitivity analysis.

Table 1

Model inputs.

Parameters	Base-case value	Range of sensitivity analysis	Distribution	Reference
Clinical inputs
EGFR mutation prevalence	50.1%	42.7%-58.8%	Beta	Local
First-line chemotherapy (cisplatin-pemetrexed) Weibull parameters
Progression-free survival
Scale parameter (λ)	0.0572	Fixed	-	[13]
Shape parameter (γ)	1.2883	Fixed	-	[13]
Overall survival
Scale parameter (λ)	0.0057	Fixed	-	[14]
Shape parameter (γ)	1.4499	Fixed	-	[14]
Odds ratios versus first-line chemotherapy			Triangular
Progression-free survival
Afatinib	8.31	3.19–23.21		[16]
Ertotinib	6.57	2.74–15.1		[16]
Gefitinib	6.12	2.87–13.5		[16]
Overall survival
Afatinib	1.11	0.62–1.96		[16]
Ertotinib	1.08	0.67–1.76		[16]
Gefitinib	0.80	0.50–1.27		[16]
Utility inputs
Utility value			Triangular
Progression-free survival	0.80	0.77–0.84		[18]
Disease progression	0.56	0.47–0.66		[18]
Disutility value			Triangular
Nausea/vomiting	-0.048	-(0.0.016–0.80)		[17]
Diarrhea	-0.047	-(0.016–0.077)		[17]
Rash	-0.032	-(0.01–0.05)		[17]
Anemia	-0.07	-(0.04–0.1)		[17]
Neutropenic fever	-0.09	-(0.06–0.12)	Triangular	[17]
Incidence of serious adverse event
Afatinib			Beta
Nausea/vomiting	0.80%	0.64%-0.96%		[6]
Diarrhea	5.4%	4.3%-6.5%		[6]
Rash	14.2%	11.4%-17.0%		[6]
Anemia	0.40%	0.3%-0.5%		[6]
Neutropenic fever	0.40%	0.3%-0.5%		[6]
Erlotinib			Beta
Nausea/vomiting	0.0%	0.0%-0.01%		[7]
Diarrhea	1.0%	0.8%-1.2%		[7]
Rash	2.0%	1.6%-2.4%		[7]
Anemia	0%	0.0%-0.01%		[7]
Neutropenic fever	0%	0.0%-0.01%		[7]
Gefitinib			Beta
Nausea/vomiting	0.3%	0.2%-0.4%		[5]
Diarrhea	3.8%	3.0%-4.6%		[5]
Rash	3.1%	2.5%-3.7%		[5]
Anemia	2.2%	1.8%-2.6%		[5]
Neutropenic fever	0.2%	0.1%-0.24%		[5]
Cisplatin-pemetrexed			Beta
Nausea/vomiting	3.6%	2.9%-4.3%		[13]
Diarrhea	0.0%	0.0%-0.01%		[13]
Rash	0.0%	0.0%-0.01%		[13]
Anemia	6.3%	5.0%-7.6%		[13]
Neutropenic fever	18.0%	14.4%-21.6%		[13]
Cost inputs (USD)	Median	IQR	Triangular	Local cohort
EGFR (per test)	324	284–432
Cost per month (including drug and non-drug costs)
Progression-free survival state
Afatinib therapy	1870	1741–2749
Erlotinib therapy	447	318–1326
Gefitinib therapy	1341	1212–2220
Cisplatin-pemetrexed therapy	2374	1159–4410
Off-chemotherapy follow up	152	122–183
Disease progression state
Palliative care	493	233–1013
Cost of serious adverse event per episode
Nausea/vomiting	160	159–177
Diarrhea	153	152–153
Rash	155	154–155
Anemia	654	653–693
Neutropenic fever	4479	2267–4729

EGFR: Epidermal growth factor receptor.

The prevalence of EGFR mutation-positive in advanced NSCLC Hong Kong patients (50.1%; range 42.7%-58.8%) was approximated from 5-year pathological data (2013–2017) of the Queen Elizabeth Hospital, largest public hospital (1,844 beds) of the Hospital Authority. The probabilities of progression-free survival and overall survival among patients treated with cisplatin-pemetrexed were estimated from the Kaplan-Meier survival curves of progression-free survival and overall survival, respectively. Data were first extracted from the corresponding Kaplan-Meier survival curves of cisplatin-pemetrexed treatment group in prospective clinical trials on EGFR mutation-positive NSCLC treatment [13, 14]. Weibull curves were used to fit to the extracted data and to estimate scale parameter (λ) and shape parameter (γ). The probability of transition at the time point of t with cisplatin-pemetrexed treatment was estimated by the following formula [15]:

A network meta-analysis, including 16 phase III randomized trials with 2,962 advanced NSCLC patients, integrated the treatment outcomes of gefitinib, erlotinib, afatinib and first-line chemotherapy [16]. The odds ratios (and 95%CI) for progression-free survival and overall survival with each TKI (gefitinib, erlotinib, afatinib) versus first-line chemotherapy reported in the network meta-analysis were adopted in the present model. The monthly probabilities of each TKI to remain in progression-free survival and in overall survival were estimated by the corresponding odds ratio and the monthly probability of event with cisplatin-pemetrexed.

Utility inputs

The utility inputs were shown in Table 1. The QALYs expected by each subject was estimated from cumulative subject-time spent in a heath state and the health state-specific utility value. The health states included progression-free survival, disease progression and death. The utility of the progression-free survival was further adjusted with disutility of treatment-related serious adverse events (SAEs) (≥Grade 3). The utility and disutility values were retrieved from literature on advanced NSCLC-related quality of life and health economic analysis [17, 18]. The incidence of SAEs with afatinib, erlotinib, gefitinib, and cisplatin-pemetrexed were adopted from the findings of randomized clinical trials on treatment of advanced NSCLC in Asian patients [5–7, 13]. The QALY gained over the model time horizon was discounted by an annual rate of 3%.

Cost inputs

The health economic analysis was conducted on direct medical costs from the perspective of Hong Kong public healthcare provider. Direct costs included biopsy-based EGFR mutation testing (for TKI arms), cost per month in progression-free survival state, cost per month for palliative care in disease progression state, and cost of managing SAEs. The healthcare resource utilization during progression-free survival and disease progression were estimated retrospectively at the Hospital Authority of Hong Kong. The Hospital Authority is the sole public healthcare provider in Hong Kong. Medical record review was conducted for patients aged ≥18 years with diagnosis of advanced (stage IIIB/IV) NSCLC (n = 400; 59% male; mean age 67±12 years) who were treated with first-line chemotherapy or TKI in 2013 to 2017 at two general hospitals (Queen Elizabeth Hospital and United Christian Hospital) of the Hospital Authority (Kowloon Central cluster and Kowloon East cluster, respectively). The study protocol was approved by the Research Ethics Committee for Kowloon Central / Kowloon East cluster. Healthcare resource utilization (including usage of drugs, clinic visit, hospitalization and diagnostic testing) were collected to estimate monthly direct medical costs for progression-free survival state and disease progression state, and management cost per episode of treatment-related SAEs, using unit cost of 2020. The costs accumulated over the 10-year model timeframe were discounted with an annual rate of 3%.

Cost-effectiveness and sensitivity analyses

Cost-effectiveness and sensitivity analyses were conducted by TreeAge Pro 2020 (TreeAge Software Inc., Williamstown, MA) and Microsoft Excel 2016 (Microsoft Corporation, Redmond, WA, USA). Base-case analysis compared the expected direct medical cost and QALYs of each EGFR mutation-guided TKI therapy with the empirical chemotherapy, assuming 100% TKI compliance as indicated by mutation test results. A treatment strategy was dominated when it gained lower QALYs at higher cost than another option, and the dominated option was eliminated from further cost-effectiveness analysis. If a treatment strategy gained more QALYs at higher cost than another alternative, incremental cost per QALY gained (ICER) of the more effective strategy was calculated: Δcost/ΔQALYs. The World Health Organization (WHO) recommended that ICER less than 1× gross domestic product (GDP) per capita to be highly cost-effective and less than 3× GDP per capita to be cost-effective [19]. The GDP per capita of Hong Kong was USD47,812 in 2019 [20] and USD143,436 (3× GDP per capita) was adopted as the willingness-to-pay (WTP) threshold in the base-case analysis. A treatment alternative was preferred if (1) it was effective in saving QALYs at lower cost, or (2) it was effective in saving QALYs at higher cost and the ICER was less than the WTP threshold.

Sensitivity analyses were performed by one-way sensitivity analysis on all model inputs to identify threshold values of influential factors. The probabilistic sensitivity analysis was performed using Monte Carlo simulation. The direct cost and QALYs of each study arm were recalculated 10,000 times by randomly drawing each of the model input from the probability distribution specified in Table 1. The probability of each strategy to be accepted as cost-effective was determined over a range of WTP from 0–200,000 USD/QALY in the acceptability curves. The incremental costs and incremental QALYs of EGFR mutation-guided TKI versus chemotherapy comparator were presented in scatter plots.

Base-case analysis

Expected direct medical costs, progression-free survival months, overall-survival months, and quality-adjusted life-years (QALYs) are shown in Table 2. In the base-case analysis of incremental costs and QALYs (Table 3), EGFR mutation-guided use of all three TKIs (afatinib, erlotinib, gefitinib) gained higher QALYs than empirical chemotherapy with cisplatin-pemetrexed. When comparing to empirical chemotherapy (as common comparator), EGFR mutation-guided erlotinib gained higher QALYs with cost-saving and the ICER of afatinib was lower than WTP threshold (143,436 USD/QALY). Both strategies of EGFR mutation-guided erlotinib and afatinib were accepted as cost-effective versus empirical chemotherapy. EGFR-guided gefitinib gained higher QALY than empirical chemotherapy at an ICER (428,208 USD/QALY) exceeding WTP, and was therefore not accepted as cost-effective.

Table 2

Expected direct medical cost, progression-free survival (PFS) months, overall-survival (OS) months, and quality-adjusted life-years (QALYs) in base-case analysis.

Strategy	Direct costs (USD)	PFS (months)	OS (months)	QALY
Empirical chemotherapy (cisplatin-pemetrexed)	21,458	8.67	33.30	1.6358
EGFR mutation-guided erlotinib	18,724	15.78	34.18	1.8072
EGFR mutation-guided gefitinib	25,997	14.30	30.98	1.6464
EGFR mutation-guided afatinib	35,808	16.79	34.50	1.8388

Table 3

Base-case analysis of increment cost, incremental QALYs and incremental cost per QALY gained (ICER).

	versus empirical chemotherapy			versus the next less costly strategy
Strategy	Incremental cost (USD)	Incremental QALYs	ICER (USD per QALY)	Incremental cost (USD)	Incremental QALYs	ICER (USD per QALY)
Empirical chemotherapy (cisplatin-pemetrexed)	-	-	-	2,734	-0.1714	Dominated* by EGFR mutation-guided erlotinib
EGFR mutation-guided erlotinib	-2,734	0.1714	Dominating empirical therapy	-	-	-
EGFR mutation-guided gefitinib	4,539	0.0106	428,208	7,273	-0.1608	Dominated* by EGFR mutation-guided erlotinib
EGFR mutation-guided afatinib	14,350	0.2030	70,690	17,084	0.0316	540,633

*Dominated strategy are eliminated from ICER analysis.

QALY: quality-adjusted life-year.

Two strategies (empirical chemotherapy and EGFR mutation-guided gefitinib) gained lower QALYs at higher costs than the erlotinib group, and were therefore dominated by the erlotinib strategy (and the dominated arms were eliminated). Comparing with EGFR mutation-guided erlotinib (as the-less-costly strategy), the afatinib strategy gained additional QALYs at higher cost by an ICER of 540,633 USD/QALY (>WTP threshold), and was not accepted to be cost-effective. EGFR mutation-guided erlotinib was the only cost-effective strategy versus both the common comparator and the next-less-costly strategy in the base-case analysis.

Sensitivity analysis

One-way sensitivity analysis found that the base-case results were robust to the variation of all model inputs and no threshold value was identified. The EGFR mutation-guided afatinib strategy gained the highest QALYs with the ICER (versus EGFR mutation-guided erlotinib) exceeding the WTP threshold. The monthly cost of afatinib therapy was further examined in an extended one-way sensitivity analysis from the base-case value (USD1870) to lower limit (USD200), for identification of a threshold value. The ICER of EGFR mutation-guided afatinib became lower than the WTP threshold (and accepted as cost-effective) when the monthly cost of afatinib therapy was lower than USD818. The compliance to use TKI (base-case value 100%, range 50%-100%) among EGFR mutation-positive patients was also added to one-way sensitivity analysis, and no threshold value was identified.

The probabilistic sensitivity analysis was performed by 10,000 Monte Carlo simulations. Comparing to empirical chemotherapy, the EGFR mutation-guided afatinib gained additional 0.2727 QALYs (95%CI 0.2699–0.2756; p<0.01) by incremental cost of USD18,838 (95%CI 18,733–18,944; p<0.01); erlotinib gained additional 0.2217 QALYs (95%CI 0.2194–0.2241; p<0.01) with cost saving of USD281 (95%CI 218–345; p<0.01); gefitinib gained additional 0.0487 QALYs (95%CI 0.0468–0.0507; p<0.01) by incremental cost of USD6,937 (95%CI 6,868–7,005; p<0.01). Figs 2–4 showed the scatter plots of the increment cost and QALYs gained by each EGFR mutation-guided TKI versus empirical chemotherapy. EGFR mutation-guided afatinib, erlotinib and gefitinib were accepted as cost-effective (with QALY gains and ICER<WTP) versus empirical chemotherapy in 86.2%, 97.2%, and 49.5% of the 10,000 simulations.

Fig 2

Scatter plot of incremental cost against QALY gained by EGFR mutation-guided afatinib versus empirical chemotherapy.

QALY: quality-adjusted life-year; WTP: willingness-to-pay.

Fig 3

Scatter plot of incremental cost against QALY gained by EGFR mutation-guided erlotinib versus empirical chemotherapy.

QALY: quality-adjusted life-year; WTP: willingness-to-pay.

Fig 4

Scatter plot of incremental cost against QALY gained by EGFR mutation-guided gefitinib versus empirical chemotherapy.

QALY: quality-adjusted life-year; WTP: willingness-to-pay.

Comparing EGFR mutation-guided afatinib versus erlotinib, afatinib gained additional 0.0510 QALYs (95%CI 0.0474–0.0546; p<0.01) by incremental cost of USD19,120 (95%CI 19,030–19,209). Fig 5 showed the scatter plots of EGFR mutation-guided afatinib versus erlotinib. The afatinib strategy was accepted as cost-effective (with QALY gains and ICER<WTP) in 31.2% of the time comparing to erlotinib.

Fig 5

Scatter plot of incremental cost against QALY gained EGFR mutation-guided afatinib versus EGFR mutation-guided erlotinib.

QALY: quality-adjusted life-year; WTP: willingness-to-pay.

To examine the acceptability of four treatment arms simultaneously, the probabilities of each treatment strategy to be accepted as cost-effective were shown in the acceptability curves over a wide range of WTP (0–200,000 USD/QALY) (Fig 6). EGFR mutation-guided afatinib, erlotinib, gefitinib and empirical chemotherapy were accepted to be preferred strategy in 0%, 98%, 0% and 2% of time at WTP 47,812 USD/QALY (1x GDP per capita), and in 30%, 68%, 2% and 0% of time at WTP 143,436 USD/QALY (3x GDP per capita), respectively.

Fig 6

Acceptability curves of empirical chemotherapy, EGFR mutation-guided afatinib, erlotinib, and gefitinib for treatment of advanced NSCLC to be cost-effective against willingness-to-pay.

EGFR: epidermal growth factor receptor; NSCLC: non-small-cell lung cancer.