Pesticide exposure and lung cancer risk: A case-control study in Nakhon Sawan, Thailand

Background: Pesticide exposure might increase risk of lung cancer. The purpose of this study was to investigate the association between the historical use of pesticides commonly found in Thailand, and lung cancer. Methods: This case-control study compared a lifetime pesticide exposure of 233 lung cancer cases, and 447 healthy neighbours matched for gender, and age (±5 years). Data on demographic, pesticide exposure and other related factors were collected using a face-to-face interview questionnaire. Associations between lung cancer and types of pesticides as well as individual pesticides were analyzed using logistic regression adjusted for gender (male, female), age (≤54, 55-64, 65-74, ≥75), cigarette smoking ( never smoked, smoked < 109,500, smoked ≥ 109,500), occupation (farmer, non-farmer), cooking fumes exposure (yes, no), and exposure to air pollution (yes, no). Results: It was found that lung cancer was positively associated with lifetime use of herbicides and insecticides. Compared to people in the nonexposed groups, those in Q3-Q4 days of using herbicides and insecticides had an elevated risk of lung cancer, with odds ratio (OR) between 2.20 (95% confidence interval (CI) 1.24-3.89), and 3.99 (95% CI 1.62-7.11) (p < 0.001). For individual pesticides, those presenting a significant association with lung cancer were dieldrin (OR = 2.56; 95% CI 1.36-4.81), chlorpyrifos (OR = 3.29; 95 % CI 1.93-5.61), and carbofuran (OR = 2.10; 95% CI 1.28-3.42). It was also found, for the first time, carbofuran, glyphosate, and paraquat to be strongly associated with lung cancer. Conclusions: The results showed that lung cancer among Thai people in Nakhon Sawan province is associated with previous pesticide use. In addition to dieldrin and chlorpyrifos, we also found carbofuran, glyphosate, and paraquat to be strongly associated with lung cancer. These issues should receive more attention since these chemicals are used widely.


Introduction
Lung cancer is a common and deadly type of cancer.In 2018, there were 2.1 million people around the world diagnosed with lung cancer, and 1.8 million died of the disease 1 .In 2018, Thailand had 170,495 incidences, and 114,199 deaths of lung cancer 2 .Besides genetic factors 3 , a major risk factor of lung cancer is cigarette smoking 4,5 .However, lung cancer was also related to other risk factors, including asbestos, crystalline silica, radon, polycyclic aromatic hydrocarbons, diesel engine exhaust particles, chromium, and nickel 6,7 .Previous studies have also linked cooking fumes to lung cancer 8 .
Pesticide exposure might also cause lung cancer 9 .The association between pesticides and lung cancer were presented around 50 years ago among grape farmers 10 .A large study in the United States found that lung cancer cases increase with the number of years working as a licensed pesticide applicator 11 .
Another study in USA reported an increased risk of lung cancer among acetochlor herbicide users (RR = 1.74, 95% CI 1.07-2.84) 12.In Pakistan, a study also found a strong association between pesticide exposure and lung cancer (OR = 5.1, 95% CI 3.1-8.3) 13.
Some studies can also link individual pesticides to lung cancer.In the USA, a study evaluated 50 pesticides and found that seven-dicamba, metolaclor, pendimethalin, carbofuran, chlorpyrifos, diazinon, and dieldrin-to be positively associated with lung cancer 14 .Another study also showed a significantly increased risk of lung cancer among applicators who had been exposed to dieldrin 15 .Jones et al. 16 reported an increased lung cancer incidence among male pesticide applicators with the highest exposure category of diazinon (odds ratio (OR) = 1.6, 95% confidence interval (CI) 1.11-2.31).Other individual pesticides that had been associated with lung cancer were chlopyrifos 17 , diazinon 18 , and pendimenthalin 19 .
To our knowledge, the association between lung cancer and pesticides has never been studied before among Thai people.The objective of this study was to investigate associations between pesticide exposure and lung cancer among people living in Nakhon Sawan province, Thailand.The results can be used for the prevention of lung cancer, and to support the global literature.
From 299 living cases registered during the study period, 32 died during the year, 20 cases were in stage IV (undifferentiated) cancer, and the other 14 not willing to participate in the study.After exclusion of those cases, 233 (participation rate = 77.9%)were contacted in person, and participated in this study.From 233 cases, 126 were confirmed by Computerized Tomography scan (CT scan)/ Magnetic Resonance Imaging (MRI)/ ultrasound of the whole abdomen/ Chest X-ray (CXR), 62 by histology of primary or metastasis, 21 by cytology of haematology, and 24 by history and physical exam.
Controls were neighbours who did not have lung or any other cancer, but were of the same gender, and age (±5 years) as the cases.In each case, two controls were selected by the interviewer using convenience sampling.In this study, data from 458 controls were used as a comparison group.
The minimum sample size was determined to be 229 for cases and 458 for controls using Kelsey's formula 21 (unmatched population base case-control study).The assumptions used were as follows: proportion of case with pesticide exposure was 0.5 22 , proportion of control with exposure was 0.4 23 , and the ratio of case to control was 1:2 24 .

Questionnaire
Data on pesticide exposure and other risk factors were collected using a questionnaire previously used in a study on pesticide exposure and diabetes 25 .The questionnaire has two major parts (provided as Extended data in English) 26 .Part 1 is about demographic data.We collected data on gender, age, marital status, education, occupation, living duration in the community, distances between home and farmland, exposure to air pollution (i.e., cooking smoke, working in a factory with air pollution; asbestos, diesel engine exhaust, silica, wood dust, painting and welding exposure).Data on smoking status, number of cigarettes smoked per day, and the total number of years smoked was also collected.Number of cumulative cigarettes smoked over a life time was calculated by multiplying the number of cigarettes smoked per day by the number of years.Those who smoked <109,500 cigarettes were considered a light smoker, while those who smoked ≥109,500 cigarettes were a heavy smoker 27 .
In Part 2, information on the historical use (mix or spray) of pesticides were collected.In this study, pesticides were categorized into five groups: insecticides (organochlorine, organophosphate, carbamate, and pyrethoid), herbicides, fungicides, rodenticides, and molluscicides.For each groups

Amendments from Version 3
The discussion section was updated.More information from the Agriculture Health Study by 33 was added.
Any further responses from the reviewers can be found at the end of the article of pesticides, we collected data on the numbers of years and days sing pesticides.The data of lifetime pesticide exposure days were then computed by multiplying the total years of exposure by the number of days per year.This study also collected data on the use of 35 individual pesticides commonly found in Thailand.
Pesticide exposure data were collected by the researcher and two village health volunteers working full-times for the project.Both case and control were interviewed by the same interviewer.Prior to data collection, all interviewers were trained on how to interview and properly use the questionnaire.

Statistical analysis
Collected data were analysed using IBM SPSS Statistics (version 25) and OpenEpi (version 3.5.1).P values <0.05 were considered statistically significant.Demographic data was analyzed using descriptive statistics.The associations were determined between lung cancer and groups of pesticides (herbicides, insecticides, fungicides, and molluscicides), between lung cancer and 17 individual compounds.Data were analyzed with conditional and unconditional logistic regression but the results were similar, and thus, only the results from the unconditional logistics regression are reported.Both crude and adjusted ORs with 95% confidence intervals (CIs) were presented.Adjusted ORs were analyzed using multiple logistic regressions controlled for gender (male, female), age (≤54, 55-64, 65-74, and ≥75), cigarette smoking (never smoked, smoked < 109,500, smoked ≥ 109,500), occupation (farmer, non-farmer), cooking fumes exposure (yes, no), and exposure to air pollution i.e., working in factories with air pollution) (yes, no).In addition to the fundamental confounding factors, variables with statistically difference between cases and controls were included in a regression model.
Cumulative exposure days on groups of pesticides were categorized into either quartiles (Q1-Q4; Q1 being the lowest exposure and Q4 the highest) or tertile (T1-T3), depending on the number of subjects in each group.The lung cancer risk was then predicted, using nonexposed group as a reference.

Ethical considerations
This study was approved by the Ethics Board of Naresuan University (project number 550/60).Written informed consent was obtained from each subject before the interviewing process.

Demographic information
In this study, most of study participants were male with a mean age of around 65.Both cases and controls have similar gender, age, marital status, education, occupation, period of residence, distances, pollution exposure, and cigarette smoking.More detailed demographic data among case and control groups were in Table 1 and in Underlying data 28 .

Lung cancer and pesticide exposure
After adjusting for confounding factors, lung cancer was positively associated with historical exposure of study participants to herbicides, insecticides and fungicides (Table 2).

Discussion
The study results showed a positive association between lung cancer and the historical use of herbicides and insecticides (Table 3).Herbicides and insecticides have a stronger association to lung cancer than fungicides.Compared to the nonexposed group, the Q1 group demonstrated lower risks of lung cancer than those in Q2 and Q3, who showed elevated risks, while Q4 had the highest risk, (Q4) (OR = 3.99, 95% CI 1.62-7.11).A similar pattern was also observed among the users of insecticides.The risk of lung cancer exponentially increased due to extended periods of using insecticides, (Q3-Q4) with OR between 2.20 (95% CI 1.24-3.89)and 2.24 (95% CI 1.33-3.72).The highest category of years using herbicides and insecticides also showed a positive association with lung cancer (OR = 1.71; 95% CI 1.33-1.53and OR=1.82; 95% CI 1.05-3.16).
These results were consistent with literature indicating the potential carcinogenicity of pesticides 29 .In an experimental study, exposure to pesticides caused the production of reactive oxygen species (ROS), an oxygen-containing species containing an unpaired electron, such as superoxide, hydrogen peroxide, and hydroxyl radical, which are highly unstable and may cause DNA damage, protein damage, mutagenicity, necrosis, and apoptosis 30 .Pesticides may also increase the risk of cancer via other mechanisms including genotoxicity, tumour promotion, epigenetic effects, hormonal action and immunotoxicity 31 .In epidemiological study, evidence linked pesticide exposure to lung cancer are increasing, and the issue will be further discussed in the following section.
For individual pesticides, the study found lung cancer to be statistically associated with dieldrin (OR = 2.56; 95 % CI 1.36-4.81),chlorpyrifos (OR = 3.29; 95% CI 1.93-5.61),and carbofuran (OR = 2.10; 95% CI 1.28-3.42)(Table 3).Dieldrin is an extremely persistent organic pollutant linked to many health problems, e.g., Parkinson's disease, breast cancer, affecting the immunity system, the reproductive, and nervous systems 32 .In the USA, the Agricultural Health Study (AHS) found seven pesticides including dicamba, metolaclor, pendimethalin, carbofuran, chlorpyrifos, diazinon, and dieldrin to be positively associated with lung cancer 14 .Further studies found dieldrin exposure to relate to the highest tertile of days use (RR = 5.30; 95% CI 1.50-18.60) 15.After fifteen years of follow-up, the latest AHS by Bonner et al. 33 18 .Similar results were also replicated in later studies of the AHS cohort for chlorpyrifos (RR = 1.80; 95% CI 1.00-3.23),which are referring to applicators in the lowest category of exposure 17 .However, the last updated results from the AHS did not support the association 33 .At this time, chlorpyrifos still not banned by Thai government.On the other hand, chlorpyrifos was the primary insecticide imported to Thailand (1,193,302 kilograms in 2013) 34 .This study also found three more pesticides, e.g.carbofuran, glyphosate, and paraquat which are significantly associated with lung cancer.To our knowledge, this study is the first to report the association.Carbofuran (2,3-dihydro-2,2-dimethylbenzofuran-7-yl methylcarbamate) is one of the most toxic broad-spectrum carbamate insecticides.
In laboratory studies, although evidence on carcinogenicity is inconclusive, carbofuran has been demonstrated to have mutagenic properties; and there are studies that have linked it to other types of cancer, e.g.lymphoma in mice 3 .In the Agricultural Health Study, lung cancer risk of carbofuran was extensively studied but no significant association to lung cancer was found 33,35 .In Thailand, a study reported on 87 different commercial brands of insecticides which were used on 202 rice fields in Suphanburi Province (abamectin 40%, followed by chlorpyrifos 30%, and carbofuran 20%), 93 brands of plant hormones, and 56 brands of chemicals for the control of plant diseases 36 .
Although, a rough estimation of the use of glyphosate and paraquat were not significantly associated with lung cancer, more detail exposure data showed that people in the higher category of cumulative exposure day, especially those in Q3 and Q4, had an elevated risk of lung cancer (P for trend <0.001) (Table 3).Paraquat (dipyridylium), also known as Gramoxone, is a non-selective herbicide commonly used worldwide 33,37 .
Although, the mechanism of toxicity has not been clearly defined, research found that paraquat can cause damage to the lungs, kidneys, and the liver 38 .At the cellular level, paraquat can produce reactive oxygen species (ROS), the superoxide free radical, which can initiate or promote carcinogenesis 39 .In a case study, pulmonary fibrosis was found in the patient with paraquat poisoning 40 .In Thailand, paraquat was banned in May, 2020.
Glyphosate [N-(phosphonomethyl)glycine, also known as roundup] is another broad-spectrum herbicide that has been used widely in Thailand and other countries 33,37 .In the past decade, cancer potency of glyphosate received much attention and several comprehensive literature reviews are available 41,42 .
In 2015, WHO revised carcinogenicity of glyphosate and classified it as "probable carcinogen" 42 .Laboratory studies found glyphosate to increase incidence of tumour, chromosomal damage, and oxidative stress 42 .Evidence from epidemiological studies were limited.Although it was previously reported to relate to non-Hodgkin lymphoma, glyphosate did not relate to cancer at any sites in the large perspective Agricultural Health Study 33,43,44 .
The potential limitations of this type of study were the recall bias where cases and controls can recall past exposure differently.Cases tend to memorize exposure better, particularly when they know or are aware of what caused their illness 45 .
However, with limited available information on the issues in Thailand, we did not expect participants to be aware of pesticides as causal factor for lung cancer.It was very likely that the study could have exposure misclassification due to the participants not being able to recall or name the pesticides they used in the past.In addition, data on pesticide exposure were obtained solely from the interview questionnaire without any exposure measurement.However, this information bias usually occurs evenly across a case and control group, and only has a negative effect on the association 46 .Selection bias might also occur when using non-random sampling or when the study sample did not represent the study population.However, in this study, data from all of the lung cancer patients except those who were severely ill, were collected to minimize the bias.

Conclusion
This study found that the occurrence of lung cancer among people in Nakhon Sawan province, Thailand is associated with pesticide use.Out of 17 individual pesticides investigated, dieldrin, chlorpyrifos, and carbofuran showed significant associations with incidence of lung cancer.These results are consistent with the literature from other parts of the world.This study found that carbofuran, glyphosate, and paraquat to be strongly associated with lung cancer.These issues should receive more attention since these chemicals have been used widely.Further studies should focus on identifying more individual pesticides that could cause lung cancer, as well as other types of cancer.
This project contains the following underlying data: • Dataset_pesticide and lung cancer (SAV and CSV).(All underlying data gathered in this study.) • Data Dictionary (DOCX).
This project contains the following extended data: • Questionnaire-pesticide and lung cancer Thailand (DOCX).
(Study questionnaire in English.) Data are available under the terms of the Creative Commons Zero "No right reserved" data waiver (CC0 1.0 Public domain dedication).
Yes, we agree that the conclusion was inconsistent with the discussion.Therefore, it was amended as follows:

Conclusion
This study found that the occurrence of lung cancer among people in Nakhon Sawan province, Thailand is associated with pesticide use.Out of 17 individual pesticides investigated, three insecticides (dieldrin, chlorpyrifos, and carbofuran), and two herbicides (glyphosate, and paraquat) were associated with incidences of lung cancer.These results were moderately supported by the literature.More studies are still required to confirm the results and to identify more individual pesticides that could cause lung cancer, as well as other types of cancer.These issues should receive more attention since these chemicals have been used widely.
Competing Interests: No competing interests were disclosed.
Author Response 17 Sep 2021

Chudchawal Juntarawijit
Comment: The conclusion that the significant associations observed for chlorpyrifos and carbonfuran are consistent with the literature from other parts of the world seems to be inaccurate and inconsistent with the early discussion of these chemicals.

Response:
Would you please clarify the point and suggest which statement should be revised and how to properly state it.The characterization of matching on weak confounding factors as "loose-matching" seems to overlook the bias introduced by matching on confounders.As Neil Pearce states: "In essence, the matching process makes the controls more similar to the cases not only for the matching factor but also for the exposure itself.This introduces a bias that needs to be controlled in the analysis" (Pearce N. Analysis of matched case-control studies.BMJ (2016)). 1 This is a well-recognized consequence of matching in case-control studies and mitigating this introduced bias in the analysis is fundamental to conducting a matched case-control study.The sparse data problem that is emphasized in the response is a common problem for most studies regardless of whether matching procedures were used.The use of conditional logistic regression can help with both these issues and should not be ignored because of a presumption of "loosematching."A comparison between the proportional hazards risk estimates with the unconditional logistic regression risk estimates seems to indicate that little if any bias was introduced by matching.This is likely due to a weak association between the matching factors and exposure.The methods section should state that conditional and unconditional logistic regression were used, that the results were similar, and results from the unconditional logistics regression are reported.

2) Response:
In this study, we actually collected data from 35 individual pesticides, but 17 of them were excluded due to small sample size (less than 5 in each cell).Therefore, the OR groups may be larger than the individual OR ones.
This issue is unresolved, and depending on which table or exposure metric, one can render different conclusions about the association between pesticides and lung cancer.In table 2, herbicide days of use has a strong monotonic exposure response gradient, but in table 3, none of the reported herbicides have an association with lung cancer.

Additional Comment:
This is a well-recognized consequence of matching in case-control studies and mitigating this introduced bias in the analysis is fundamental to conducting a matched case-control study.The sparse data problem that is emphasized in the response is a common problem for most studies regardless of whether matching procedures were used.The use of conditional logistic regression can help with both these issues and should not be ignored because of a presumption of "loose-matching."A comparison between the proportional hazards risk estimates with the unconditional logistic regression risk estimates seems to indicate that little if any bias was introduced by matching.This is likely due to a weak association between the matching factors and exposure.The methods section should state that conditional and unconditional logistic regression were used, that the results were similar, and results from the unconditional logistics regression are reported.

Response:
The statement that both conditional and unconditional analyses were performed has been added to the method section.

Comment:
2) Response: In this study, we actually collected data from 35 individual pesticides, but 17 of them were excluded due to small sample size (less than 5 in each cell).Therefore, the OR groups may be larger than the individual OR ones.
This issue is unresolved, and depending on which table or exposure metric, one can render different conclusions about the association between pesticides and lung cancer.In table 2, herbicide days of use has a strong monotonic exposure response gradient, but in table 3, none of the reported herbicides have an association with lung cancer.

Response:
Taking the suggestion, we decided to recheck the data and found some individual pesticides to have enough sufficient participants to calculate cumulative exposure days.Reanalysis of the data using quartile of exposure days, found good results.Those exposed to glyphosate and paraquat in Q4 and Q3 had a significant association with lung cancer.So, the issue was solved.Thank you very much for your thoughtful suggestions.Without it, we would miss this important finding.
Also, to make it more consistent, we decided to use 'nonexposed' as a reference instead of Q1, and the reanalysis of the data.This change caused only a minor change in the results.
All parts of the paper (abstract, results, and discussion section) were updated.

Comment: Additional Comment:
1) The discussion section on chlorpyrifos and carbofuran describes results from the Agricultural Health Study.However, cited reports are not the most recent reports on these pesticides and lung cancer.A more recent report from 2017, found no association with either chlorpyrifos or carbofuran.The discussion should be updated in light of this more recent report (https://ehp.niehs.nih.gov/doi/10.1289/EHP456). 2 well, but crucial information regarding several specific details are missing from the report.These details, and other concerns, are described below.

Comments:
Crucial information about the lung cancer cases is missing.Specifically, were the cases comprised of 1 st primary lung cancer or were lung cancer cases with a prior history of another cancer, including lung, eligible to participate in the study.Were the lung cancer cases' diagnosis histologically confirmed?What was the stage and grade of these lung cancer cases?On average, how long after their diagnosis were lung cancer cases interviewed? 1.
Were potential controls excluded if they had a prior history of cancer? 2.
The methods state that two neighbor controls were selected randomly.This implies that there a sampling frame of some sort.That sampling frame for random selection needs to be adequately described.

3.
The controls were matched on age and sex to the cases.This necessitates a conditional logistic regression to account for the selection bias introduced by matching.Unconditional logistic regression is inappropriate for a matched case-control study.Breaking the matching and adjusting for the matching factors may not bias the odds ratios, but this should be confirmed by comparing ORs estimated with unconditional logistic regression and conditional logistic regression.

4.
In table 2, there is striking qualitative confounding for the pesticide classes (yes vs. no).For instance, the crude OR organophosphates is 0.63 (95% CI = 0.46-0.87)while the adjusted OR is 1.77 (95% CI = 1.22-2.57).The use of unconditional logistic regression might explain this as the crude estimate for such a regression is inappropriate.That notwithstanding, a number of other variables were included in the regressions.Given this qualitative confounding, additional analyses to identify variables or combination of variables is driving this confounding is warranted.

5.
Tobacco smoking is a recognized strong risk factor for lung cancer and a known potential confounder in studies of other exposures and lung cancer.As such, substantial efforts to mitigate confounder are often employed.In this study, smoking was a binary (ever vs. never) variable that may not adequately capture the interrelationship between pesticide use and lung cancer to control confounding.More detailed smoking information, if available, should be explored to determine the potential for residual confounding of the reported associations 6.
In table 1, smoking is not associated with lung cancer.This suggests that selection forces in the recruitment of cases and controls are biasing the study results.It seems unusual that 61% of lung cancer cases were never smokers.Is this a typical feature of lung cancer in Thailand?

7.
The results reported in tables 2 and 3 seem to be internally inconsistent.For instance, the ORs for organophosphates depicted in table 4 indicate a strong association with lung cancer with days of use (Q4 vs. Q1 OR = 28.43 (95% CI = 11.11-72.76);an extremely large 8. magnitude.However, the ORs for specific organophosphate insecticides are much more modest, although the statistically significant associations with chlorpyfos and dielrin.A similar pattern is evident for herbicide as well.This lack of internal consistency really points to the methodological limitations as a likely explanation for the observed association.
Recall bias was discussed as a limitation, but nothing is mentioned about other threats to internal validity.For instance, the potential for selection bias to arise from the recruitment strategies.As mentioned above, the lack of an association with smoking seems to indicate something is awry.In addition, exposure misclassification is undoubtedly present and should be discussed in the Discussion along with the other potential limitations.9.
References 20 and 35 are the same report.10.

Are the conclusions drawn adequately supported by the results? Partly
Competing Interests: No competing interests were disclosed.
Reviewer Expertise: Cancer epidemiology, pesticides, air pollution, occupational and environmental epidemiology I confirm that I have read this submission and believe that I have an appropriate level of expertise to state that I do not consider it to be of an acceptable scientific standard, for reasons outlined above.
another cancer, including lung, eligible to participate in the study.Were the lung cancer cases' diagnosis histologically confirmed?What was the stage and grade of these lung cancer cases?On average, how long after their diagnosis were lung cancer cases interviewed?

Response:
The cases comprised of 1 st primary lung cancer.The cases were confirmed by Computerized Tomography scan (CT scan), Magnetic Resonance Imaging (MRI), ultrasound of the whole abdomen, and chest radiography or Chest X-ray (CXR), and histology of primary and metastasis.More information was added to Table 1, and provided in the Table below.
On average, the patients were interviewed approximately1 year after they had been diagnosed with lung cancer.

Response:
Yes, potential controls were excluded if they had a prior history of cancer.

Comment:
3. The methods state that two neighbor controls were selected randomly.This implies that there a sampling frame of some sort.That sampling frame for random selection needs to be adequately described.

Response:
The control was, in fact, selected using convenience sampling.The information in the manuscript has been revised.

Comment:
4. The controls were matched on age and sex to the cases.This necessitates a conditional logistic regression to account for the selection bias introduced by matching.Unconditional logistic regression is inappropriate for a matched case-control study.Breaking the matching and adjusting for the matching factors may not bias the odds ratios, but this should be confirmed by comparing ORs estimated with unconditional logistic regression and conditional logistic regression.

Response:
Matching of a few variables can be considered loose-matching, therefore, it is more appropriate to analyze using unconditional logistic regression.Kuo and team (2018) said that "There is a presumption that matched data need to be analyzed by matched methods.
Conditional logistic regression has become a standard for matched case-control data to tackle the sparse data problem.The sparse data problem, however, may not be a concern for loose-matching data when the matching between cases and controls are not unique, and one case can be matched to other controls without substantially changing the association.Data matched on a few demographic variables are clearly loose-matching data, and we hypothesize that unconditional logistic regression is a proper method to perform."(Kuo, Duan & Grady, 2018)* We gained interesting information by analyzing some of the data using Cox regression.The comparison between unconditional analysis and the Cox regression, yielded similar results.(see Table below).Comparing OR between Cox regression and logistic regression).

Comment:
5. In table 2, there is striking qualitative confounding for the pesticide classes (yes vs. no).For instance, the crude OR organophosphates is 0.63 (95% CI = 0.46-0.87)while the adjusted OR is 1.77 (95% CI = 1.22-2.57).The use of unconditional logistic regression might explain this as the crude estimate for such a regression is inappropriate.That notwithstanding, a number of other variables were included in the regressions.Given this qualitative confounding, additional analyses to identify variables or combination of variables is driving this confounding is warranted.

Response:
After re-categorizing the smoking variable, and correcting a mistake on the variable coding, the new analysis yielded more consistent results with crude OR at 1.35 (95%CI 0.98-1.86)and adjusted OR at 1.40 (95%CI 0.97-2.02).
6. Tobacco smoking is a recognized strong risk factor for lung cancer and a known potential confounder in studies of other exposures and lung cancer.As such, substantial efforts to mitigate confounder are often employed.In this study, smoking was a binary (ever vs. never) variable that may not adequately capture the interrelationship between pesticide use and lung cancer to control confounding.More detailed smoking information, if available, should be explored to determine the potential for residual confounding of the reported associations We actually collected data on the amount of cigarettes and smoking duration of the study participants, and then the information was used to compute number of cigarettes smoked by them in their life time.After grouping smoking status into "never smoked", "smoked <109500 cigarettes", and "smoked ≥109500 cigarettes", a significant difference between case and control was found.The data was then used for the analysis of the odds ratio.

Comment:
7. In table 1, smoking is not associated with lung cancer.This suggests that selection forces in the recruitment of cases and controls are biasing the study results.It seems unusual that 61% of lung cancer cases were never smokers.Is this a typical feature of lung cancer in Thailand?

Response:
Yes, 61% of the cases never smoked is acceptable.It was reported that smoking prevalence of Thai males decreased from 60% to 39%, and from 5% to 2.1% in females between 1991 and 2014 [1].While a survey in 2017 reported a smoking prevalence of 20.7% of the total adult population over 15 years old [2].
It was interesting to note that in this study, 49.2% of the cases were adenocarcinoma lung cancer which has a limited relation to cigarette smoking, whereas squamous cell, and small cell lung carcinoma are highly related to smoking [3, 4].
to the methodological limitations as a likely explanation for the observed association.

Response:
In this study, we actually collected data from 35 individual pesticides, but 17 of them were excluded due to small sample size (less than 5 in each cell).Therefore, the OR groups may be larger than the individual OR ones.

Comment:
9. Recall bias was discussed as a limitation, but nothing is mentioned about other threats to internal validity.For instance, the potential for selection bias to arise from the recruitment strategies.As mentioned above, the lack of an association with smoking seems to indicate something is awry.In addition, exposure misclassification is undoubtedly present and should be discussed in the Discussion along with the other potential limitations.

Response:
The problems of selection bias and exposure misclassification has been further discussed in the manuscript as suggested.The problem of lack of association with smoking has already been solved.

Lyon, France
Dear authors, I was pleased to review your paper that describes a case-control study in Thailand including 233 incident lung cancer cases and 458 controls focusing on exposures to pesticides.Please find enclosed my comments for your consideration.In the: Introduction, first paragraph, I think you mean "Polycyclic aromatic hydrocarbons"?I would not call a paper from 1999 "Recent studies….." because it's >20 years old.
Methods, first paragraph, it's a case-control study, not a case-controlled study; it would be good to include a few more details such as any time limit for having resided in the province?; from where did the TCB receive cases?; were the diagnosis confirmed by some diagnostic tool?Please clarify if the 299 were contacted and 229 accepted, or if only 229 were contacted and accepted.I would be surprised if the latter, and wonder why the other were not contacted.We also wish to know the "participation rate" among the control subjects.Neighbours are not a random sample it's a convenience sample.If you mean that the interviewer randomly selected control subjects among all neighbours you need to explain how this was done, e.g.within a distance from the house or "snowball" technique.Why do you adjust for farming (yes/no)?Please explain your rational.It does not make sense to me.
Questionnaire, the English questionnaire does not indicate that the number of days of pesticide use is per year, so it seems strange that lifetime exposure is calculated by multiplying years with days.Please also clarify if "exposure" refers to "personally mix or apply pesticides" only, or if it also includes working in the fields?Provide more details regarding the data collection e.g. were the interviewers employed for the study full-time, or were they students?, were there any quality control measures implemented, e.g.double interviews of a proportion of subjects, were the interviewers interviewing both cases and controls?Results, it is very strange that there is not difference between cases and controls regarding smoking, if you have an explanation for this please discuss it later.
Discussion, I don't think that "the association were closer for herbicides and insecticides", possibly "stronger" or "more pronounced", and I prefer "more days" rather than "higher days".
Among the limitations I think there is more to information bias, e.g. it is commonly difficult to assess exposure to specific chemicals because people don't know the names or don't recognize exposure.I must admit that I get suspicious that there are no missing in the data and no category for "don't know" in the questionnaire.I would add potential selection bias to the discussion; although we don't really know the participation rate among controls or how neighbours were selected, they are generally not an ideal control population.

Disclaimer:
Where authors/reviewers are identified as personnel of the International Agency for Research on Cancer / World Health Organization, the authors/reviewers alone are responsible for the views expressed in this article and they do not necessarily represent the decisions, policy or views of the explain how this was done, e.g.within a distance from the house or "snowball" technique.Why do you adjust for farming (yes/no)?Please explain your rational.It does not make sense to me.

Response:
The mistake was corrected and more information on residency, TCB, and diagnostic confirmation was added to the Methods.The information of the number of cases was clarified; and information on the participation rate was also provided.
For the question, why did we adjust for farming?Actually, we tried to adjust for occupations since it is very likely that they will be exposed to pesticides differently.At first, there were several types of occupations, but due to the small number of participants in each category, the groups were limited to "farmer" and "none-farmer".These two groups tended to have different risks of exposure to environmental pesticides, due to the nature of their work and physical health.

Comment:
Questionnaire, the English questionnaire does not indicate that the number of days of pesticide use is per year, so it seems strange that lifetime exposure is calculated by multiplying years with days.Please also clarify if "exposure" refers to "personally mix or apply pesticides" only, or if it also includes working in the fields?Provide more details regarding the data collection e.g. were the interviewers employed for the study full-time, or were they students?, were there any quality control measures implemented, e.g.double interviews of a proportion of subjects, were the interviewers interviewing both cases and controls?

Response:
More information was added and the mistakes were corrected.In this study, "exposure" refers to "personally mixed and/or applied pesticides" only, not working in the field.More information of interviewers was added to the methods.There were no other quality control measures implemented.
Comment: Results, it is very strange that there is not difference between cases and controls regarding smoking, if you have an explanation for this please discuss it later.Response: Data on cigarette smoke was reanalyzed and the difference was observed using a new category.

Comment:
Discussion, I don't think that "the association was closer for herbicides and insecticides", possibly "stronger" or "more pronounced", and I prefer "more days" rather than "higher days".

Response:
The term "closer" was changed to "stronger" Comment: Among the limitations, I think there is more to information bias, e.g. it is commonly difficult to assess exposure to specific chemicals because people don't know the names or don't recognize exposure.I must admit that I get suspicious that there are no missing in the data and no category for "don't know" in the questionnaire.I would add potential selection bias to the discussion; although we don't really know the participation rate among controls or how neighbours were selected, they are generally not an ideal control population.

Response:
Yes, we agree that it was likely that some of participants could not recall or know the name of the pesticides used.If this type of bias occurs it would be equal between both the case and control groups, and minimize the association between exposure to pesticides and lung cancer.More information regarding bias was added to the Discussion section, and more information about the control group was added to the Methods.Those who could not recall or "don't know" the name of the pesticides were categorized as "not used".
Competing Interests: No competing interests were disclosed.
The benefits of publishing with F1000Research: Your article is published within days, with no editorial bias • You can publish traditional articles, null/negative results, case reports, data notes and more • The peer review process is transparent and collaborative • Your article is indexed in PubMed after passing peer review • Dedicated customer support at every stage • For pre-submission enquiries, contact research@f1000.com

1
Department of Epidemiology and Environmental Health, School of Public Health and Health Professions, University at Buffalo, Buffalo, NY, USA 2 Department of Epidemiology and Environmental Health, School of Public Health and Health The sparse data problem, however, may not be a concern for loose-matching data when the matching between cases and controls are not unique, and one case can be matched to other controls without substantially changing the association.Data matched on a few demographic variables are clearly loosematching data, and we hypothesize that unconditional logistic regression is a proper method to perform."(Kuo, Duan & Grady, 2018)*

Table .
More information on morphology and stage of the study cases.