Figures

Self-reported cancer diagnosis. Note patients may indicate more than one cancer.

Self-reported race/ethnicity. Note patients may report more than one race/ethnicity.

Self-reported genes. Note patients may report more than one gene.

Self-reported genes among Stomach cancer patients. Note patients may report more than one gene.

Self-reported gastric lesions. Note patients may report more than one lesion.

Table 1. Overall demographics and health characteristics

Characteristic	N = 1,632¹
Age
Less than 29	43 (2.6%)
30-39	173 (11%)
40-49	373 (23%)
50-59	394 (24%)
60-69	410 (25%)
70 and over	239 (15%)
Sex
Female	1,552 (95%)
Male	80 (4.9%)
Race_Ethnicity
American Indian, Alaska Native, or First Nations	1 (<0.1%)
Asian or Asian American	24 (1.5%)
Black or African-American	10 (0.6%)
Hispanic White	19 (1.2%)
Hispanic, Latino, or Latinx	38 (2.3%)
Mixed race/ethnicity	22 (1.3%)
Non-Hispanic White	1,502 (92%)
Other	16 (1.0%)
Any_Gastric_Lesion	798 (49%)
H_pylori	100 (6.1%)
Fam_Hist_Stomach_Cancer
Don't know	272 (17%)
No	957 (59%)
Yes	403 (25%)
Stomach_Cancer	50 (3.1%)
¹ n (%)

Table 2. Risk factors by any gastric lesion

Any grastric lesion includes any of the following:

Helicobacter Pylori (H. pylori)
Gastritis
Gastric Ulcer
Gastro-esophageal reflux (GERD)
Intestinal Metaplasia
Barrett’s Esophagus
Pernicious Anemia

HR genes: ATM, BARD1, BRCA1, BRCA2, BRIP1, CHEK2, PALB2, RAD51C, RAD51D, NBN

Lynch genes: MLH1, MSH2, MSH6, PMS2, EPCAM

Hereditary Diffuse Gastric Cancer (HDGC) genes: CDH1, CTNNA1

Juvenile Polyposis Syndrome genes: SMAD4, BMPR1A

Peutz-Jeghers Syndrome gene: STK11 (also known as LKB1)

Li-Fraumeni Syndrome gene: TP53

Recessive cancer predisposition syndrome genes: MUTYH, NTHL1

All other genes not included: CDKN2A, ENG, NF1, PTEN

Table 3. Risk factors by H. pylori status

Table 4. Cancers by family history of gastric cancer

Note: n=272 who responded “Don’t know” are not included in table 4.

Table 5. Risk factors by gastric cancer status

Characteristic	Overall N = 782¹	No N = 732¹	Yes N = 50¹	p-value
Fam_Hist_Stomach_Cancer				<0.001²
Don't know	117 (15%)	111 (15%)	6 (12%)
No	463 (59%)	445 (61%)	18 (36%)
Yes	202 (26%)	176 (24%)	26 (52%)
Any_Gastric_Lesion				0.024²
No	402 (51%)	384 (52%)	18 (36%)
Yes	380 (49%)	348 (48%)	32 (64%)
H_pylori				0.009³
No	743 (95%)	700 (96%)	43 (86%)
Yes	39 (5.0%)	32 (4.4%)	7 (14%)
HR				<0.001²
No	178 (23%)	148 (20%)	30 (60%)
Yes	604 (77%)	584 (80%)	20 (40%)
Lynch				>0.9²
No	668 (85%)	625 (85%)	43 (86%)
Yes	114 (15%)	107 (15%)	7 (14%)
HDGC				<0.001³
No	731 (93%)	701 (96%)	30 (60%)
Yes	51 (6.5%)	31 (4.2%)	20 (40%)
Anti_acid_ever				>0.9²
No	326 (42%)	305 (42%)	21 (42%)
Yes	456 (58%)	427 (58%)	29 (58%)
Anti_acid_time				<0.001³
No	326 (42%)	305 (42%)	21 (43%)
Daily	166 (21%)	146 (20%)	20 (41%)
1-4 times weekly	115 (15%)	115 (16%)	0 (0%)
1-4 times monthly	44 (5.6%)	42 (5.7%)	2 (4.1%)
1-4 times in past 6 months	102 (13%)	99 (14%)	3 (6.1%)
Other/As needed	28 (3.6%)	25 (3.4%)	3 (6.1%)
Missing	1	0	1
¹ n (%)
² Pearson’s Chi-squared test
³ Fisher’s exact test

Let’s look at cell counts for (gastic cancer/no cancer), H pylori, and HR

##                              HR  No Yes
## Stomach_ORNO_Cancer H_pylori           
## No                  No          144 556
##                     Yes           4  28
## Yes                 No           27  16
##                     Yes           3   4

Table 6. Risk factors by (gastric cancer or any gastric lesion)

Cases: Gastric cancer, or any gastric lesion

Controls: (Non-gastric cancer and no family history of gastric cancer) or no cancer

Note we are assuming comparability of groups.

Characteristic	Overall N = 1,200¹	0 N = 384¹	1 N = 816¹	p-value
H_pylori				<0.001²
No	1,100 (92%)	384 (100%)	716 (88%)
Yes	100 (8.3%)	0 (0%)	100 (12%)
Anti_acid_ever				<0.001²
No	417 (35%)	226 (59%)	191 (23%)
Yes	783 (65%)	158 (41%)	625 (77%)
HR				<0.001²
No	298 (25%)	65 (17%)	233 (29%)
Yes	902 (75%)	319 (83%)	583 (71%)
Lynch				0.002²
No	976 (81%)	332 (86%)	644 (79%)
Yes	224 (19%)	52 (14%)	172 (21%)
HDGC				0.011²
No	1,137 (95%)	373 (97%)	764 (94%)
Yes	63 (5.3%)	11 (2.9%)	52 (6.4%)
FAP				0.7²
No	1,178 (98%)	376 (98%)	802 (98%)
Yes	22 (1.8%)	8 (2.1%)	14 (1.7%)
Juvenile_Polyposis_Syndrome				>0.9³
No	1,199 (100%)	384 (100%)	815 (100%)
Yes	1 (<0.1%)	0 (0%)	1 (0.1%)
Peutz_Jeghers_Syndrome				0.6³
No	1,197 (100%)	384 (100%)	813 (100%)
Yes	3 (0.3%)	0 (0%)	3 (0.4%)
Li_Fraumeni_Syndrome				0.2³
No	1,195 (100%)	384 (100%)	811 (99%)
Yes	5 (0.4%)	0 (0%)	5 (0.6%)
Recessive_Cancer_Predisposition				0.2³
No	1,186 (99%)	382 (99%)	804 (99%)
Yes	14 (1.2%)	2 (0.5%)	12 (1.5%)
Stomach_Genes				0.018²
No	36 (3.0%)	5 (1.3%)	31 (3.8%)
Yes	1,164 (97%)	379 (99%)	785 (96%)
Other_genes
No	1,200 (100%)	384 (100%)	816 (100%)
¹ n (%)
² Pearson’s Chi-squared test
³ Fisher’s exact test

Table 7. Associations between cancer-predisposing gene groups and risk for (gastric cancer or any gastric lesion).

Models are adjusted for sex and current age category. 95% CI uses the profile likelihood method.

Gene	No cancer/Non-gastric cancer & No family history n=384	Gastric/Lesion n=816	OR (95% CI)	P-Value
HRYes	319 (83.07)	583 (71.45)	0.42 (0.3-0.58)	0.0000002
LynchYes	52 (13.54)	172 (21.08)	1.83 (1.29-2.62)	0.0008615
HDGCYes	11 (2.86)	52 (6.37)	2.79 (1.46-5.79)	0.0031809
FAPYes	8 (2.08)	14 (1.72)	1.07 (0.44-2.76)	0.8871420
Recessive_Cancer_PredispositionYes	2 (0.52)	12 (1.47)	2.69 (0.69-17.95)	0.2124594
Stomach_GenesYes	379 (98.7)	785 (96.2)	0.25 (0.08-0.6)	0.0047796
BRCA1Yes	81 (21.09)	134 (16.42)	0.69 (0.5-0.96)	0.0247509
BRCA2Yes	138 (35.94)	193 (23.65)	0.54 (0.41-0.71)	0.0000118
CDH1Yes	11 (2.86)	50 (6.13)	2.67 (1.4-5.57)	0.0048704
H_pyloriYes	0 (0)	100 (12.25)	20163339.96 (83.72-2.26938371154776e+70)	0.9645141

Interaction effects - (Unable to do, 0 cell)

Parameter	Odds Ratio	SE	95% CI	z	p
HR (Yes)	0.38	0.06	(0.28, 0.53)	-5.68	< .001
H pylori (Yes)	9.66e+06	8.93e+09	(0.00, Inf)	0.02	0.986
HR (Yes) × H pylori (Yes)	2.65	2682.28	(0.00, Inf)	9.65e-04	> .999

Table 8. Stomach Cancer: Interaction effects of HR and HDGC genes with H. pylori any gastric lesion

Models adjusted for sex and entry age category. 95% CI uses the profile likelihood method. Note that interaction analysis is likely underpowered due to low n=50 stomach cases.

8a. HR x H. pylori

##              Stomach_ORNO_Cancer  No Yes
## HR  H_pylori                            
## No  No                           144  27
##     Yes                            4   3
## Yes No                           556  16
##     Yes                           28   4

Parameter	Odds Ratio	SE	95% CI	z	p
HR (Yes)	0.14	0.05	(0.07, 0.26)	-5.88	< .001
H pylori (Yes)	3.63	2.98	(0.65, 18.30)	1.58	0.115
HR (Yes) × H pylori (Yes)	1.44	1.46	(0.19, 10.95)	0.36	0.720

8b. HR x any gastric lesion

##                        Stomach_ORNO_Cancer  No Yes
## HR  Any_Gastric_Lesion                            
## No  No                                      65  13
##     Yes                                     83  17
## Yes No                                     319   5
##     Yes                                    265  15

Parameter	Odds Ratio	SE	95% CI	z	p
HR (Yes)	0.07	0.04	(0.02, 0.20)	-4.73	< .001
Any Gastric Lesion (Yes)	0.95	0.40	(0.42, 2.18)	-0.13	0.900
HR (Yes) × Any Gastric Lesion (Yes)	3.35	2.26	(0.92, 13.38)	1.79	0.073

8c. HDGC x H. pylori

##               Stomach_ORNO_Cancer  No Yes
## HDGC H_pylori                            
## No   No                           669  24
##      Yes                           32   6
## Yes  No                            31  19
##      Yes                            0   1

Parameter	Odds Ratio	SE	95% CI	z	p
HDGC (Yes)	18.95	7.29	(8.94, 40.71)	7.65	< .001
H pylori (Yes)	5.48	2.78	(1.88, 14.10)	3.36	< .001
HDGC (Yes) × H pylori (Yes)	6.50e+05	5.74e+08	(7.58e-73, )	0.02	0.988

8d. HDGC x any gastric lesion

##                         Stomach_ORNO_Cancer  No Yes
## HDGC Any_Gastric_Lesion                            
## No   No                                     373   6
##      Yes                                    328  24
## Yes  No                                      11  12
##      Yes                                     20   8

Parameter	Odds Ratio	SE	95% CI	z	p
HDGC (Yes)	76.55	47.73	(23.61, 279.09)	6.96	< .001
Any Gastric Lesion (Yes)	4.20	1.96	(1.79, 11.50)	3.08	0.002
HDGC (Yes) × Any Gastric Lesion (Yes)	0.08	0.06	(0.02, 0.35)	-3.23	0.001

Table 9. Colorectal Cancer: Interaction effects of HR and Lynch genes with H. pylori

Models adjusted for sex and entry age category. 95% CI uses the profile likelihood method. Note that interaction analysis is likely underpowered due to low n=72 colorectal cases. There are cells with 0 value.

9a. HR x H. pylori

## , ,  = No
## 
##      
##         No  Yes
##   No   308   14
##   Yes 1161   77
## 
## , ,  = Yes
## 
##      
##         No  Yes
##   No    46    3
##   Yes   17    6

Parameter	Odds Ratio	SE	95% CI	z	p
HR (Yes)	0.08	0.02	(0.04, 0.14)	-8.41	< .001
H pylori (Yes)	1.09	0.74	(0.23, 3.72)	0.13	0.900
HR (Yes) × H pylori (Yes)	4.82	4.08	(0.96, 28.50)	1.86	0.063

9b. Lynch x H. pylori

## , ,  = No
## 
##      
##         No  Yes
##   No  1241   82
##   Yes  228    9
## 
## , ,  = Yes
## 
##      
##         No  Yes
##   No    19    2
##   Yes   44    7

Parameter	Odds Ratio	SE	95% CI	z	p
Lynch (Yes)	13.74	3.99	(7.89, 24.81)	9.01	< .001
H pylori (Yes)	1.50	1.13	(0.23, 5.35)	0.54	0.592
Lynch (Yes) × H pylori (Yes)	2.52	2.38	(0.45, 20.67)	0.98	0.325

##                                                  H_pylori   No  Yes
## Race_Ethnicity                                                     
## American Indian, Alaska Native, or First Nations             1    0
## Asian or Asian American                                     19    5
## Black or African-American                                   10    0
## Hispanic White                                              19    0
## Hispanic, Latino, or Latinx                                 29    9
## Mixed race/ethnicity                                        18    4
## Non-Hispanic White                                        1421   81
## Other                                                       15    1

Locations of the 41 NHWs that have H. pylori

## 
##          California            Colorado         Connecticut             Florida 
##                   7                   2                   2                   5 
##             Georgia                Iowa           Isla Wake              Kansas 
##                   1                   1                   1                   2 
##            Kentucky               Maine            Maryland           Minnesota 
##                   1                   2                   1                   2 
##            Missouri             Montana       New Hampshire          New Jersey 
##                   1                   1                   1                   3 
##            New York      North Carolina        North Dakota                Ohio 
##                   4                  11                   1                   3 
##              Oregon Prefer not to share      South Carolina               Texas 
##                   2                   8                   1                   6 
##                Utah            Virginia          Washington       West Virginia 
##                   1                   1                   6                   1 
##           Wisconsin 
##                   3

Exploring the outcome and exposures

Considering Dr. Nuno’s supervised learning with penalized logistic regression (LASSO)

## [1] 0.008249971

## 23 x 1 sparse Matrix of class "dgCMatrix"
##                     s1
## (Intercept)  1.3298938
## Q4          -1.1754926
## Q5          -3.1477437
## Q6          -1.2546994
## Q7           .        
## Q8          -3.3529903
## Q10          3.2172535
## Q12          .        
## Q13         -2.1302120
## Q15          .        
## Q16          .        
## Q17         -0.3752814
## Q18         -1.7291057
## Q19         -2.1982152
## Q20          .        
## Q21         -2.3113357
## Q22         -4.6991217
## Q23          .        
## Q24          .        
## Q25          3.2175091
## Q26          .        
## Q27          .        
## Q28          0.9061810

## 23 x 1 Matrix of class "dgeMatrix"
##                       s1
## (Intercept)  3.780641719
## Q4           0.308666905
## Q5           0.042948925
## Q6           0.285161566
## Q7           1.000000000
## Q8           0.034979597
## Q10         24.959474354
## Q12          1.000000000
## Q13          0.118812107
## Q15          1.000000000
## Q16          1.000000000
## Q17          0.687095927
## Q18          0.177443022
## Q19          0.111001101
## Q20          1.000000000
## Q21          0.099128758
## Q22          0.009103269
## Q23          1.000000000
## Q24          1.000000000
## Q25         24.965855270
## Q26          1.000000000
## Q27          1.000000000
## Q28          2.474853062

## 0.8677273

## 
## Call:
## glm(formula = gc_outcome ~ ., family = "binomial", data = df2[, 
##     c("gc_outcome", c(selected_vars, "age", "Sex"))])
## 
## Coefficients:
##               Estimate Std. Error z value Pr(>|z|)    
## (Intercept)  -16.90047 5594.88585  -0.003 0.997590    
## Q4No           1.76954    1.24400   1.422 0.154894    
## Q5No           4.02067    0.93945   4.280 1.87e-05 ***
## Q6No          15.21325 3956.18051   0.004 0.996932    
## Q8No           4.12965    0.81485   5.068 4.02e-07 ***
## Q10No        -19.22340 3956.18051  -0.005 0.996123    
## Q13No          3.05290    0.82703   3.691 0.000223 ***
## Q17No          1.73746    1.51635   1.146 0.251871    
## Q18No          2.73895    0.94947   2.885 0.003918 ** 
## Q19No          3.56111    0.98727   3.607 0.000310 ***
## Q21PALB2      -4.12984    1.32272  -3.122 0.001795 ** 
## Q22PMS2      -19.40782 1368.06309  -0.014 0.988681    
## Q25RAD51D     19.79616 3956.18051   0.005 0.996008    
## Q28TP53       14.72877 3956.18048   0.004 0.997030    
## age            0.04091    0.02353   1.739 0.082100 .  
## SexMale        1.18301    0.92644   1.277 0.201619    
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## (Dispersion parameter for binomial family taken to be 1)
## 
##     Null deviance: 180.71  on 137  degrees of freedom
## Residual deviance: 104.72  on 122  degrees of freedom
## AIC: 136.72
## 
## Number of Fisher Scoring iterations: 16

##  (Intercept)         Q4No         Q5No         Q6No         Q8No        Q10No 
## 4.573173e-08 5.868136e+00 5.573857e+01 4.046030e+06 6.215628e+01 4.481101e-09 
##        Q13No        Q17No        Q18No        Q19No     Q21PALB2      Q22PMS2 
## 2.117661e+01 5.682874e+00 1.547077e+01 3.520226e+01 1.608542e-02 3.726410e-09 
##    Q25RAD51D      Q28TP53          age      SexMale 
## 3.956958e+08 2.492430e+06 1.041760e+00 3.264197e+00

Characteristic	OR	95% CI	p-value
Q4
APC	—	—
No	5.87	0.51, 78.8	0.2
Q5
ATM	—	—
No	55.7	10.7, 472	<0.001
Q6
BARD1	—	—
No	4,046,030	0.00,	>0.9
Q8
BRCA1	—	—
No	62.2	14.5, 374	<0.001
Q10
BRIP1	—	—
No	0.00		>0.9
Q13
CHEK2	—	—
No	21.2	4.64, 124	<0.001
Q17
MLH1	—	—
No	5.68	0.19, 118	0.3
Q18
MSH2	—	—
No	15.5	2.63, 117	0.004
Q19
MSH6	—	—
No	35.2	5.99, 309	<0.001
Q21
No	—	—
PALB2	0.02	0.00, 0.16	0.002
Q22
No	—	—
PMS2	0.00		>0.9
Q25
No	—	—
RAD51D	395,695,778	0.00,	>0.9
Q28
No	—	—
TP53	2,492,430	0.00,	>0.9
age	1.04	1.00, 1.09	0.082
Sex
Female	—	—
Male	3.26	0.52, 21.2	0.2
Abbreviations: CI = Confidence Interval, OR = Odds Ratio

Unsupervised

##         y
## clusters  0  1
##        1 62 47
##        2 26  3

Cancer cases do not cleanly separate and are scattered in multiple regions. Some mild grouping but no distinct clusters. Mutation profiles do not strongly separate gastric cancer vs non-cancer patients in a linear low-dimensional structure (PCA). Clustering structure exists, but is weak and not cleanly distinct in PCA space

Principal component analysis of mutation profiles did not reveal clear separation between gastric cancer and non-cancer patients. Similarly, hierarchical clustering showed only weak group structure, suggesting that gastric cancer risk is not driven by global mutation patterns but rather by specific genetic alterations

PCA to detect global variance structure is weak/none. Clustering to see patient subgroups is weak. LASSO to detect sparse predictive features has some strong signals.

Our data suggests that there are no clear groups, but there are a few important genes. This pattern is common in biomedical data as diseases are often not cluster-driven, but are feature-driven for specific mutations.

FORCE Analysis

LCC Lab Community Team & April Vang

2026-04-30