Posted as supplied by author.

Statistical methods for web

The database has a hierarchical structure with: 1 or more findings per colonoscopy; 1 or more colonoscopies per individual; and 1 or more individuals per family. Except where clearly indicated, we have reported the most advanced neoplastic lesion on a given colonoscopic examination. The outcomes of the initial colonoscopies are presented as directly standardized proportions. The proportions of different outcomes on follow-up colonoscopies have been adjusted using multinomial logistic regression in Stata (StataCorp. 2003. Statistical Software: Release8.0 College Station, TX: Stata Corporation). Standardization and adjustment has been made for age (in 5-year groups), sex and family history, as appropriate. Where cited, p-values are based on (multinomial) logistic regression adjusting for possible confounders and taking into account the possible correlation between related individuals.

Rates of (first) advanced neoplasia were calculated using the total time between the first and last surveillance colonoscopy as denominator. Mortality rates are relative to the total time from the first colonoscopy until the end of the follow-up period (the earlier of 31/12/02 and death). Three different analyses were used to study colorectal cancer incidence in order to address the complications created by screen-detected cancers. The simplest analysis considers cancer incident between the first and last colonoscopy (for individuals with two or more colonoscopies). In this analysis we exclude cancers detected on the first colonoscopy, but include cancers detected on the last colonoscopy. The second analysis includes all individuals with at least one screening colonoscopy. We simply compare the observed cumulative-rate (or Nelson-Aalen estimate[1] (from the time of the first colonoscopy) to the expected cumulative rate censoring all follow-up at 31/12/2001. As no colorectal cancers were identified by flagging that had not already been entered in our clinical database, we also carried out our analysis of colorectal cancer diagnosed up to the end of 2003. This approach is useful for observing if and when the observed curve (with a jump due to screen-detected cases at time zero) crosses the expected curve, but it lacks statistical power if substantial numbers of screened individuals only have very limited follow-up. The final approach is to consider only those individuals with potentially three years or more follow-up (i.e. those enrolled prior to 31/12/98) and to include all cancers diagnosed prior to 31/12/01 regardless of mode of detection. Expected numbers were calculated using cancer registry rates from England for 1997 in five-year age bands for men and women separately.[2] Expected numbers of deaths were calculated using incidence rates and survival rates (for colorectal cancer diagnosed in 1997 in England) rather than from mortality rates.[3]

After calculating the expected number of cancers and cancer deaths in each family type and in each age group, using standard population rates, we applied three sets of familial relative risks corresponding to our best estimate and what we consider to be the lowest and highest reasonable estimate based on the literature (Table 5).[4][5][6][7][8][9][10][11][12] These estimates are particularly influenced by the Swedish population results of Dong and Hemminki,[12] but also take into account how the relative risk should change with age in someone who is a first degree relative of a gene carrier. Taking account of age, the average relative hazards based on the best (lowest, highest) in our cohort are 17.1 (7.9, 23.8) for HNPCC and 5.7 (2.8, 10.2) for moderate-risk familial colorectal cancer, corresponding to lifetime risk in women (to age 79) of 37% (19%, 47%) and 15% (8%, 26%) respectively. These should be compared to the observed relative risk of between 20.7 and 32.7 for individuals with an affected parent and sibling[12] and 4.0 for a first degree relative for CRC cancer under age 60.[7]

References for statistical methods

1. Marubini E, Valsecchi MG. Analysing survival data from clinical trials and observational studies. Chichester: Wiley, 1995.

2. National Statistics. Cancer statistics - registrations 1995-1997. Series MN1 no 28. The Stationery Office 2001.

3. Sasieni P. The expected number of deaths from cancer during follow-up of an initially cancer-free cohort. Epidemiology 2003;14(1):108-110. Gastroenterology 1997;112:594-642.

4. Fuchs CS, Giovannucci EL, Colditz GA, Hunter DJ, Speizer FE, Willett WC. Prospective study of family history and the risk of colorectal cancer. N Engl J Med 1994;331(25):1669-1674.

5. St. John DJ, McDermott FT, Hopper JL, Debney EA, Johnson WR, Hughes ESR. Cancer Risk in Relatives of Patients with Common Colorectal cancer. Ann Intern Med 1993;118(10):785-790.

6. Dunlop MG, Farrington SM, Carothers AD, Wyllie AH, Sharp L, Burn J, et al. Cancer risk associated with germline DNA mismatch repair mutations. Hum Molec Genetics 1997;6(1):105-110.

7. Goldar DE, Easton DF, Cannon-Albright LA, Skolnick MH. Systematic population-based assessment of cancer risk in first-degree relatives of cancer probands. J Natl Cancer Inst 1994;86:1600-1608.

8. Hall NR, Bishop DT, Stephenson BM, Finan PJ. Hereditary Susceptibility to Colorectal Cancer. Dis Col Rectum 1996;39(7):739-743.

9. Slattery ML, Kerber RA. Family history of cancer and colon cancer risk: the Utah population database. J Natl Cancer Inst 1994;86:1618-1626.

10. Vasen HFA, Wijnen JT, Menko FH, Kleibuker JH, Taal BG, Griffioen G, et al. Cancer risk in families with hereditary nonpolyposis colorectal cancer diagnosed by mutation analysis. Gastroenterology 1996;110:1020-1027.

11. Voskuil DW, Vasen HF, Kampman E, van�t Veer P. Colorectal cancer risk in HNPCC families with hereditary nonpolyposis colorectal cancer diagnosed by mutation analysis. Int J Cancer 1997;72:205-209.

12. Dong C, Hemminki K. Modification of cancer risks in offspring by sibling and parental cancers from 2,1112 616 nuclear families. Int J Cancer 2001;92:144-150.

�

Posted as supplied by author.

Table A (Supplementary) Relative risks of CRC incidence corresponding to the lowest, best and highest reasonable estimate based on the literature, by age group and family history

Age group	Extent
	HNPCC			3FDR			2FDR			1FDR
	L	M	H	L	M	H	L	M	H	L	M	H
<25	20	47	60	6	19	24	3	8	12	2	6.5	13
25-29	20	47	60	6	19	24	3	8	12	2	6.5	13
30-34	15	39	45	6	16	24	3	7	10	2	5.5	12
35-39	15	34	45	6	14	24	3	6	9	2	5	11
40-44	15	29	45	6	12	24	2.5	5.5	8	1.8	4.5	10
45-49	10	26	30	5	11	20	2.5	5	8	1.8	4	9
50-54	10	22	30	4	10	16	2.5	4.5	7	1.8	3.5	8
55-59	10	19	30	4	9	16	2	4	6	1.6	3	7
60-64	8	16	24	4	8	16	2	3.5	6	1.6	3	6
65-69	5	13	15	3	7	12	2	3	5	1.4	2.5	6
70-74	5	10	15	3	6	12	2	3	5	1.4	2	5
75-79	3	7	9	3	6	12	1.5	2.5	4	1.2	2	4
80-84	3	7	9	3	6	12	1.5	2.5	4	1.2	2	4
85+	3	7	9	3	6	12	1.5	2.5	4	1.2	2	4