Cancer Causes & Control

, Volume 19, Issue 5, pp 469–479

Accuracy of self-reported family history of cancer in a large case–control study of ovarian cancer

Authors

  • Marie Soegaard
    • Institute of Cancer EpidemiologyDanish Cancer Society
  • Allan Jensen
    • Institute of Cancer EpidemiologyDanish Cancer Society
  • Kirsten Frederiksen
    • Institute of Cancer EpidemiologyDanish Cancer Society
  • Estrid Høgdall
    • Institute of Cancer EpidemiologyDanish Cancer Society
  • Claus Høgdall
    • The Gynaecologic ClinicThe Juliane Marie Centre, Rigshospitalet
  • Jan Blaakær
    • Department of Gynaecology and ObstetricsAarhus University Hospital
    • Institute of Cancer EpidemiologyDanish Cancer Society
    • The Gynaecologic ClinicThe Juliane Marie Centre, Rigshospitalet
Original Paper

DOI: 10.1007/s10552-007-9108-3

Cite this article as:
Soegaard, M., Jensen, A., Frederiksen, K. et al. Cancer Causes Control (2008) 19: 469. doi:10.1007/s10552-007-9108-3

Abstract

Objective

To evaluate the reliability of self-reported family history of cancer in first-degree female relatives and to examine possible determinants of accurate reporting.

Methods

Women with ovarian cancer and controls were recruited between 1995 and 1999 and interviewed. The study comprised 579 cases and 1,564 controls with 6,265 first-degree female relatives. Self-reported familial cancer diagnoses were validated from registry data. Sensitivity, specificity, and kappa were calculated, and possible determinants were examined by logistic regression.

Results

The sensitivity of self-reporting ranged from 0.78 to 0.90 for all cancers but was lower for self-reporting of most site-specific cancers, ranging from 0.29 to 0.94. The specificity of self-reporting ranged from 0.91 to 0.99 for cancer in general and from 0.99 to 1.00 for site-specific cancers. Type of relative, age at interview, and length of education influenced the sensitivity and specificity significantly. The odds ratio for ovarian cancer was higher when based on registry data than on self-reported data and was significant (OR = 2.58 vs. 1.56).

Conclusions

Cancer diagnoses in first-degree relatives are not always accurately reported by patients with ovarian cancer or by controls. The results indicate that studies of associations with family cancer history should validate self-reported family cancer diagnoses as carefully as possible.

Keywords

Hereditary cancerFamily cancer historyValidationOvarian cancerHereditary breast and ovarian cancer

Introduction

Many epidemiological studies of family cancer history are based on self-reported data, and it is important to determine how reliable these data are in order to interpret such studies. For counseling at the individual level, the accuracy of self-reported family cancer history is also important, as decisions about surveillance or prophylactic surgery depend on the family history of cancer in cases in which the BRCA mutation status is unknown. Although it is regarded as standard practice to confirm as many diagnoses as possible when counseling the families of cancer patients, it can be difficult to obtain verification of the various diagnoses, especially in parts of the world where registries are incomplete or inexistent or hospital records are difficult to trace.

Various types of studies have been conducted to determine the reliability of self-reported data on family history of cancer [116]. The majority of these are case-only studies making them unable to elucidate whether case–control status influences the accuracy of reporting. Of the six case-control studies published [13, 1215], only four included both positive and negative self-reports; although when interpreting self-reported family history for clinical purposes, specificity is just as important as sensitivity.

A family history of ovarian cancer is one of the strongest identified risk factors for ovarian cancer [17], as women with a single first-degree relative affected by ovarian cancer have a threefold increase in the relative risk for ovarian cancer [18]. Epidemiological data and genetic analyses clearly indicate that mutations in one of the breast and ovarian cancer susceptibility genes, BRCA1 and BRCA2, are responsible in almost all families in which multiple cases of ovarian and breast cancer have clustered [19, 20].

The purpose of the study reported here was to evaluate the reliability of self-reported family cancer history in first-degree female relatives by ovarian cancer patients and control women and to examine the possible determinants of accurate reporting. To our knowledge, this is the largest case-control study of ovarian cancer on reliability of self-reported cancer in female first-degree relatives. We found it important to report our data as the study is population-based, identification of first-degree relatives is almost complete and is not limited to a certain time period as we went manually through the church books, if information on relatives could not be found in the registries. Linkage to the Danish Cancer Registry secures completeness of the familial cancer diagnoses and furthermore, we were able to trace cancer diagnoses further back than most other studies.

Materials and methods

The study is based on data from a large Danish population-based case–control study of ovarian cancer, the MALOVA study. A detailed description of the study population and the data collected has been published previously [21].

Cases

Briefly, the case group consisted of women aged 35–79 years, who were recruited between January 1995 and May 1999 from 16 gynecological departments around Denmark. Women scheduled for explorative laparotomy or laparoscopy on the suspicion of an ovarian tumor were asked to participate in the study, with blood and tissue samples, and a personal interview. In order to ensure that all eligible cases in the study area were included, we linked the study database to the Danish Cancer Registry every second month. Women registered in the Cancer Registry with ovarian cancer but not initially included in the study were contacted by letter and asked to participate in an interview.

Written informed consent was obtained from each participant before the interview, which were conducted by female nurses, medical students, and laboratory technicians who were trained by the same physician and given written guidelines on how to perform the interview. The questionnaire was designed for use in personal interviews and was used to ascertain social, reproductive, medical, gynaecological, and dietary histories. A life-event calendar, specially designed for this study, was used to obtain detailed information on pregnancies, abortions, births, breastfeeding periods, contraception, fertility drug use, and hormone therapy. Information was collected on female first-degree relatives, and the index person was requested to report the name and date of birth of each relative, whether they had had cancer and, if they had, in which organ and at what age it was diagnosed.

A total of 681 women (79.9% of those approached) with ovarian cancer agreed to participate in the study. Of these, 579 completed the personal interview, making them eligible for the analysis.

Controls

Controls were recruited simultaneously with the cases, from the same area as the case women. A random sample was drawn from the general female population aged 35–79 years of age in the study area by means of the computerized Civil Registration System, which contains the unique personal identification numbers of all inhabitants of Denmark. They were frequency matched in 5-year intervals by using the age distribution of women with ovarian cancer (1987–1992) registered in the Danish Cancer Registry. A total of 3,839 women were invited by letter to participate in the study as controls, with a personal interview and a blood sample. Contact could not be made with 301 women, 269 women were excluded because they had undergone a bilateral oophorectomy, six women had moved out of the study area and 126 women were too ill to participate, leaving 3,137 women as eligible controls. A total of 2,116 women were enrolled, of whom 1,564 participated in a personal interview and 552 participated in a telephone interview. Those who were interviewed by telephone did not differ from the other controls with respect to number of pregnancies, length of oral contraceptive use, menopausal status or hysterectomy (data not shown). As the telephone interview was less comprehensive than the personal interview, the analysis was based exclusively on the personal interviews. Table 1 shows general characteristics for the index persons.
Table 1

General characteristics for the study population

 

Cases

Controls

No. of participants

562

1525

Median (5–95%) age at interview (years)

59 (41–76)

57 (37–75)

Length of education (years)

    7–9

267 (48%)

698 (46%)

    10+

294 (52%)

827 (54%)

Pregnancy

    Ever

481 (86%)

1432 (94%)

    Never

81 (14%)

93 (6%)

Mean duration of oral contraceptive use (years)

2.27

4.05

No. of relatives identified

    Mother

519

1418

    Sister(s)

609

1678

    Daughter(s)

481

1560

Median (5–95%) follow-up* (years)

   Mother

87 (67–108)

86 (63–106)

   Sister(s)

60 (41–82)

59 (35–80)

   Daughter(s)

34 (17–51)

33 (11–49)

* Follow-up: Date of birth to date of index person’s interview

Identification of relatives

The Civil Registration System, established on 1 April 1968, has issued a unique 10-digit personal identification number to every Danish person alive at that date or born afterwards. The numbers contain information on date of birth and gender and are used throughout Danish society, including in health registries, thus securing valid linkage among registries.

During the personal interview, the index women were asked the name and date of birth of their mother and biological sister(s) and daughter(s). We confirmed these dates and obtained the dates of birth of family members for whom we had only names or no information at all in the computerized Civil Registration System. This System, however, provides links only to the mothers and sisters of persons born from the late 1950s, whereas virtually all of our study population had been born before that time. In addition, some of the mothers would have died before 1968 and would therefore not be registered in the System. When information on first-degree relatives could not be found in this System, we contacted the National Register (Folkeregisteret) in the municipality in which the woman was born. These registers were introduced in every municipality in Denmark in 1924 and record every inhabitant in the respective municipality up to the start of the Civil Registration System in 1968. The registers consist of paper files and are located in 271 municipalities. If the information could not be found in either the Civil Registration System or the National Register, we went through the church books of the parish in which the index person and her family had lived at the time of her birth to obtain the names and birth dates of the parents.

A total of 54 relatives for whom personal identifying information could not be found were excluded from the study (16 relatives of cases and 38 relatives of controls). We also excluded 429 relatives (equally distributed proportionately between cases and controls) on whom information was not available from the interview with the index person: 149 with missing information, 169 for whom the answer to the question about cancer was ‘Don’t know’, and 111 who were not mentioned by the index person in the interview but found in our subsequent investigation. If the cancers in relatives were reported as having been diagnosed before or in 1942, when the Danish Cancer Registry was established, they were excluded from the analysis (seven cancers among relatives of the case group and 14 cancers among relatives of the control group).

A total of 6,265 female first-degree family members were identified, comprising 1,937 mothers, 2,041 daughters and 2,287 sisters. Of these, 1,609 were related to 562 women with ovarian cancer, and 4656 were related to 1,525 controls. The remaining women in the study (17 ovarian cancer cases and 39 controls) did not have relatives eligible for inclusion.

Cancers in relatives

The Danish Cancer Registry was established in 1942, and notification to the register became compulsory from 1987. The completeness and validity of the Registry were already very high before notification became compulsory, with 96–98% of all cancers registered [22, 23]. The group of relatives was linked to the Danish Cancer Registry by their personal identification numbers or, if they had died before 1 April 1968, by their date of birth and full name. We identified all cancers in the relatives occurring up to the date of the interview of the respective index person.

Classification of cancer type

Diagnoses in the Danish Cancer Registry are grouped into organ-specific sites or areas on the basis of the codes of the International Classification of Diseases 7th Revision (ICD-7) [24]. The diagnoses of the cancers in relatives reported by the index women were classified into these groups for comparison. Some of the groups reflect an area rather than a specific organ owing to poor definition of the cancer site. The following groups were used: all malignant neoplasms (ICD-7: 140–205); buccal cavity, pharynx, lip, tongue, salivary glands, mouth, nasal cavities and sinuses or larynx (ICD-7: 140–148,160–161); esophagus (ICD-7: 150); stomach (ICD-7: 151); small intestine, colon and rectum (ICD-7: 152–154); liver and gall-bladder (ICD-7: 155); pancreas (ICD-7: 157); lung (ICD-7: 162–163); breast (ICD-7: 170); uterus and cervix (ICD-7: 171–174); ovaries (ICD-7: 175); other and unspecified female genital organs (ICD-7: 176); kidney and bladder (ICD-7: 180–181); skin (190–191); eye and brain (ICD-7: 192–193); thyroid (ICD-7: 194); and bone, connective tissue, lymphatic and hematopoetic tissue, lymphoma, myeloma, and leukemia (ICD-7: 196–197,200–205). Precancerous cervical lesions (dysplasia, atypia, carcinoma in situ) were not counted as cancer diagnoses in this study. Thus, if the index person reported a cervix cancer or abdominal cancer in a relative, who was registered in the Danish Cancer Registry with a cervical precancerous lesion, the diagnosis was regarded as invalid, whereas if the index person reported the diagnosis as a non-malignant condition or did not report it at all, the statement was regarded as valid. Twenty diagnoses could not be grouped because the information on the cancer did not correspond to any group; however, none of these were ovarian cancers according to the Danish Cancer Registry.

Statistical analysis

The analyses are based on registry data and family history reported in first-degree relatives of index persons.

We measured the reliability of the interview data by calculating the sensitivity of the index person’s reports as the proportion of family members reported as having cancer among those with cancer according to the Cancer Registry. The specificity was calculated as the proportion of family members reported as not having cancer among those who did not have cancer before the date of interview according to the Cancer Registry [25].

We also estimated the overall agreement, the kappa value κ defined as κ = (o-e)/(1-e), where o is the observed agreement between the personal interview and the Danish Cancer Registry and e is the expected (just by chance) proportion of agreement [26]. Kappa can be seen as the chance-corrected proportional agreement. An interpretation of the range of kappa has been given by Landis and Koch [27]. Confidence intervals (ninety-five percent) were calculated by standard methods [26].

Logistic regression was used to test for differences in sensitivity and specificity according to case or control status, type of relative, age at interview, length of education, and years since diagnosis of cancer in the relative. The analyses were adjusted for age at interview, specified as a linear function, and the generalized estimation equations approach was used to correct for correlations within the index person. The linearity of the quantitative variables, including age, additional pregnancies, and duration of oral contraceptive use, was tested and compared with linear spline effects with knots placed at the tertiles. No significant deviations from linearity were found.

We calculated odds ratios by logistic regression for the association between family history of cancer among first-degree family members and the risk for ovarian cancer on the basis of self-reported family history and validated data from the Danish Cancer Registry in order to determine whether the two sources of information led to the same risk estimate for ovarian cancer. We adjusted for age at interview, any pregnancy, additional pregnancies, and duration of oral contraceptive use, as these variables were found previously to be significant risk factors for ovarian cancer [28].

Results

Characteristics of participants

The median age at interview was for the 562 ovarian cancer cases 59 years and for the 1525 control women 57 years (Table 1). Forty-eight percent of the cases and 46% of controls had 7–9 years of education, whereas the remaining cases and controls had 10 years or more of education. Eighty-six percent of cases and 94% of controls had ever been pregnant. The mean duration of oral contraceptive use was, respectively, 2.27 years and 4.05 years for cases and controls. All together, 519 mothers to cases were identified with a median follow-up time on 87 years and 1,418 mothers to controls with a median follow-up time on 86 years. Six hundred and nine sisters were identified to the case group with a median follow-up time on 60 years, while 1,678 sisters to controls were identified with a follow-up on 59 years. We identified 481 daughters to the case group with a median follow-up on 34 years and 1,560 daughters to the control group with a median follow-up on 33 years.

Reliability of self-reported familial cancer

The sensitivity, specificity, and kappa of self-reported family history of cancer among first-degree female relatives of patients with ovarian cancer and control women are shown in Table 2. For cancers at all sites, the sensitivity of self-reported family cancer history tended to be higher for cases than for controls with regard to both sisters and daughters, but the difference was not statistically significant; there was no difference with regard to mothers. The specificity of self-reported family cancer at all sites was between 0.91 and 0.99. Kappa values corresponded to a substantial agreement of measures, except from daughters to the control group (moderate agreement).
Table 2

Sensitivity, specificity and kappa of self-reported family history of cancer versus registered diagnosis among first-degree female relatives of patients with ovarian cancer and control subjects

Cancer type (ICD-7 code)

Self-reported familial cancer

Cancer diagnosis in Danish Cancer Registry

Registered

Not registered

Sensitivity

Specificity

Kappa

All Sites (cancer yes/no)

In any first-degree female relative

  Control

Yes

459

192

0.81 (0 78–0.84)

0.95 (0.95–0.96)

0.72 (0.69–0.75)

No

109

3,896

   

  Case

Yes

206

70

0.83 (0.79–0.88)

0.95 (0.94–0.96)

0.75 (0.70–0.79)

No

41

1,292

   

In mothers

  Control

Yes

297

92

0.82 (078–0.86)

0.91 (0.90–0.93)

0.72 (0.68–0.76)

No

64

965

   

  Case

Yes

113

32

0.81 (0.74–0.87)

0.92 (0.89–0.94)

0.71 (0.65–0.78)

No

27

347

   

In sister(s)

  Control

Yes

146

73

0.78 (0.72–0.84)

0.95 (0.94–0.96)

0.68 (0.63–0.74)

No

41

1,418

   

  Case

Yes

84

32

0.87 (0.80–0.93)

0.94 (0.92–0.95)

0.74 (0.67–0.81)

No

13

480

   

In daughter(s)

  Control

Yes

16

27

0.80 (0.62–0.98)

0.98 (0.98–0.99)

0.50 (0.35–0.65)

No

4

1,513

   

  Case

Yes

9

6

0.99 (0.71–1.00)

0.99 (0.98–1.00)

0.71 (0.51–0.92)

No

1

465

   

Site-specifica

Breast

  Control

Yes

121

22

0.94 (0.90–0.98)

1.00 (0.99–1.00)

0.89 (0.85–0.93)

No

8

4,505

   

  Case

Yes

57

12

0.89 (0.81–0.97)

0.99 (0.99–1.00)

0.85 (0.79–0.92)

No

7

1,533

   

Ovarian

  Control

Yes

13

11

0.42 (0.25–0.59)

1.00 (1.00–1.00)

0.47 (0.31–0.63)

No

18

4,614

   

  Case

Yes

14

3

0.44 (0.27–0.61)

1.00 (1.00–1.00)

0.57 (0.40–0.73)

No

18

1,574

   

Stomach

  Control

Yes

7

26

0.47 (0.21–0.72)

0.99 (0.99–1.00)

0.29 (0.12–0.46)

No

8

4,615

   

  Case

Yes

4

13

0.44 (0.12–0.77)

0.99 (0.99–1.00)

0.30 (0.07–0.53)

No

5

1,587

   

Small intestine/colon/rectum

  Control

Yes

50

22

0.69 (0.59–0.80)

1.00 (0.99–1.00)

0.69 (0.60–0.78)

No

22

4,562

   

  Case

Yes

21

10

0.70 (0.54–0.86)

0.99 (0.99–1.00)

0.68 (0.55–0.82)

No

9

1,569

   

Lung/pleura

  Control

Yes

25

17

0.66 (0.51–0.81)

1.00 (0.99–1.00)

0.62 (0.50–0.75)

No

13

4,601

   

  Case

Yes

13

10

0.81 (062–1.00)

0.99 (0.99–1.00)

0.66 (049–0.83)

No

3

1,583

   

Skin

  Control

Yes

26

21

0.29 (0.20–0.39)

1.00 (0.99–1.00)

0.37 (0.30–0.48)

No

63

4,546

   

  Case

Yes

11

7

0.37 (0.19–0.54)

1.00 (0.99–1.00)

0.45 (0.27–0.63)

No

19

1,572

   

Kidney/bladder

  Control

Yes

5

3

0.29 (0.08–0.51)

1.00 (1.00–1.00)

0.40 (0.16–0.64)

No

12

4,636

   

  Case

Yes

4

2

0.40 (0.10–0.70)

1.00 (1.00–1.00)

0.50 (0.20–0.80)

No

6

1,597

   

Bone/haematop. system

  Control

Yes

23

23

0.62 (0.47–0.78)

1.00 (0.99–1.00)

0.55 (0.42–0.68)

No

14

4,596

   

  Case

Yes

7

8

0.64 (0.35–0.92)

1.00 (0.99–1.00)

0.53 (0.30–0.77)

No

4

1,590

   

aIn any first-degree female relatives

For site-specific cancers, the sensitivity was generally much lower than that for a self-reported diagnosis of cancer at any site, except for breast cancer. The highest sensitivity was seen for breast cancer, followed by cancer in the intestinal area and lung cancer. Lower sensitivity was seen for cancers at other sites or areas. The specificity of self-reported family site-specific cancer was between 0.99 and 1.00. Agreement was highest overall for breast cancer and lowest for stomach cancer. No significant differences between cases and controls were observed in sensitivity, specificity or kappa for either all cancer or site-specific cancers.

Determinants of the reliability of self-reported familial cancer

We examined potential determinants of correct reporting of cancer in a first-degree relative (sensitivity) and of the absence of cancer in a relative (specificity) (Table 3), with analyses for all cancers and for breast and ovarian cancers separately. Neither sensitivity nor specificity was significantly related to case or control status after adjustment for age at interview. The specificity was associated with type of relative, as the index persons reported more reliably about their daughters for cancer at any site (OR, 3.44; 95% CI, 2.30–5.14) and for breast cancer (OR, 16.5; 95% CI, 2.25–121) than for other relatives. Both the sensitivity and the specificity were lower with respect to mothers when sisters were used as the reference group (ORs, 0.44–0.98), specificity being statistically significantly lower for any cancer (OR, 0.58; 95% CI, 0.45–0.75).
Table 3

Potential determinants for accurate reporting of any cancer, breast cancer, and ovarian cancer

Variable

Cancer all sites

Breast cancer

Ovarian cancer

Correct/not correct

OR

95% CI

Correct/not correct

OR

95% CI

Correct/not correct

OR

95% CI

Sensitivity

Status

    Case

206/41

1.23

0.81–1.87

57/7

0.57

0.20–1.65

14/18

0.86

0.30–2.44

    Control

459/109

1.00

 

121/8

1.00

 

13/18

1.00

 

Type of relative

    Mothers

410/91

0.85

0.58–1.26

92/10

0.44

0.13–1.54

14/24

0.56

0.18–1.73

    Sisters

230/54

1.00

 

77/5

1.00

 

13/11

1.00

 

    Daughters

25/5

1.40

0.49–3.98

9/0

 

0/1

 

Age at interview

    Per year older

 

0.97

0.95–0.99

 

0.96

0.91–1.01

 

0.97

0.93–1.02

Length of education (years)

    7–9

359/72

1.00

 

97/4

1.00

 

16/19

1.00

 

    10+

306/78

0.63

0.43–0.92

81/11

0.23

0.07–0.84

11/17

0.70

0.21–2.27

  Per year of education

 

0.90

0.81–0.99

 

0.68

0.49–0.93

 

0.94

0.70–1.26

Years since diagnosis of cancer in relative

    0–10

2,57/73

1.00

 

77/9

1.00

 

12/12

1.00

 

    11–20

1,84/41

1.34

0.86–2.09

49/3

1.90

0.53–6.86

7/7

1.00

1.00–1.00

    21+

2,24/36

2.12

1.35–3.31

52/3

2.21

0.56–8.80

8/17

0.62

0.21–1.84

    Per year since diagnosis

 

1.04

0.99–1.09

 

1.04

0.99–1.09

 

1.00

1.00–1.00

Specificity

Status

    Case

1,292/70

0.96

0.72–1.29

1,533/12

0.64

0.32–1.30

1,574/3

1.15

0.32–4.06

    Control

3,896/192

1.00

 

4,505/22

1.00

 

4,614/11

1.00

 

Type of relative

    Mothers

1,312/124

0.58

0.45–0.75

1,820/15

0.98

0.50–1.91

1,891/8

0.65

0.23–1.86

    Sisters

1,898/105

1.00

 

2,187/18

1.00

 

2,257/6

1.00

 

    Daughters

1,978/33

3.44

2.30–5.14

2,031/1

16.5

2.25–121

2,040/0

 

Age at interview

    Per year older

 

0.97

0.96–0.98

 

0.98

0.95–1.01

 

1.05

1.00–1.10

Length of education years)

    7–9

2,624/173

1.00

 

3,103/24

1.00

 

3,187/6

1.00

 

    10+

2,560/88

1.67

1.24–2.23

2,930/10

2.11

0.96–4.64

2,996/8

0.98

0.30–3.23

    Per year of education

 

1.12

1.04–1.21

 

1.17

0.97–1.41

 

0.95

0.73–1.24

All OR adjusted for age as a linear function

Age at the time of the interview affected the sensitivity and specificity of reporting, as younger women were more likely to report correctly. The odds of correct reporting decreasing by 2–4% per yearly increase in age at interview, which was significant for all cancer; however, a 5% non-significant increase in specificity was seen for reports of ovarian cancer per yearly increase in age.

A longer education (>10 years) was associated with decreased sensitivity, which was significant for all cancer (OR, 0.63; 95% CI, 0.43–0.92) and breast cancer (OR, 0.23; 95% CI, 0.07–0.84). An opposite effect was seen on specificity, with greater accuracy of reporting with longer education for all cancer (OR, 1.67; 95% CI, 1.24–2.23) and breast cancer (OR, 2.11; 95% CI, 0.96–4.64). The sensitivity decreased by 6–32% per extra year of education (ORs, 0.68–0.94), whereas the specificity increased by 12–17% per extra year of education, except for ovarian cancer (ORs, 1.12–1.17).

Time since diagnosis of a relative’s cancer led to greater sensitivity for all cancers and for breast cancer, though non-significant for breast cancer. We tried to stratify into two age groups, index persons below 60 years and 60 years or above at time of interview and found significantly increased sensitivity for 21+ years since cancer diagnosis in relatives for both groups (data not shown). Reports of ovarian cancer were less accurate when the interval between the relative’s diagnosis and the time of the interview was 21 years or more (OR, 0.62; 95% CI, 0.21–1.84). For all cancer and breast cancer, there was a non-significant increase in sensitivity of 4% per year since the relative’s cancer diagnosis.

Reports of age at diagnosis of familial cancer

We assessed the accuracy of reporting of the age at diagnosis in a relative by women who correctly reported cancer in a relative (yes/no) and for those who correctly reported breast or ovarian cancer. For cancer in a relative, correct age, defined as <4 years’ difference between the reported age and the registered age, was given by 82% of controls and 79% of cases. For breast cancer in a relative, 83% of controls and 81% of cases reported the correct age; for ovarian cancer, 77% of controls and 86% of cases reported the correct age.

Associations based on reported data and validated data

Table 4 shows estimates of the associations between a family history of all cancer or breast or ovarian cancer in first-degree relatives and the risk for ovarian cancer based on reported data and Registry data. The risks for ovarian cancer associated with a family history of any cancer or of breast cancer did not differ between reported and Registry data; however, the risk for ovarian cancer was almost twice as high and significant when a relative was reported to have had ovarian cancer in the Cancer Registry (OR, 2.58; 95% CI, 1.52–4.39) than when the cancer was reported by a case (OR, 1.56; 95% CI, 0.78–3.13). This was especially pronounced for ovarian cancer in sisters, where the OR for ovarian cancer was 1.24 (95% CI, 0.44–3.49) when based on self-reported data and 2.82 (95% CI, 1.17–6.82) when based on Registry data.
Table 4

Odds ratios (OR) and 95% confidence intervals (CI) for the association between family history of cancer and risk of ovarian cancer based on self-reported family history and on the Danish Cancer Registry respectively

Family history of cancer

Self-reported data

Danish Cancer Registry

Controls

Cases

OR (95% Cl)a

Controls

Cases

OR (95% Cl)a

Cancer (all sites)

Cancer in family

    Yes

543

219

1.09 (0.88–1.34)

503

206

1.11 (0.90–1.37)

    No

982

343

1.00

1,022

356

1.00

Cancer in mother

    Yes

389

145

0.98 (0.78–1.23)

361

140

1.03 (0.81–1.30)

    No

1,136

417

1.00

1,164

422

1.00

Cancer in sister(s)

    Yes

190

98

1.40 (1.06–1.86)

162

82

1.38 (1.02–1.87)

    No

1,335

464

1.00

1,363

480

1.00

Cancer in daughter(s)

    Yes

41

15

1.00 (0.54–1.85)

20

9

1.23 (0.55–2.75)

    No

1,484

547

1.00

1,505

553

1.00

Breast

Breast cancer in family

    Yes

140

61

1.19 (0.86–1.66)

128

58

1.21 (0.86–1.70)

    No

1,385

501

1.00

1,397

504

1.00

Breast cancer in mother

    Yes

79

28

0.94 (0.59–1.50)

74

28

0.98 (0.61–1.56)

    No

1,446

534

1.00

1,451

534

1.00

Breast cancer in sister(s)

    Yes

58

34

1.61 (1.02–2.53)

50

29

1.52 (0.94–2.48)

    No

1,467

528

1.00

1,475

533

1.00

Breast cancer in daughter(s)

    Yes

5

5

2.90 (0.82–10.2)

4

5

3.65 (0.96–13.9)

    No

1,520

557

1.00

1,521

557

1.00

Ovarian

Ovarian cancer in family

    Yes

23

14

1.56 (0.78–3.13)

31

29

2.58 (1.52–4.39)

    No

1,502

548

1.00

1,494

533

1.00

Ovarian cancer in mother

    Yes

14

8

1.71 (0.69–4.23)

21

17

2.28 (1.17–4.45)

    No

1,511

554

1.00

1,504

545

1.00

Ovarian caner in sister(s)

    Yes

10

6

1.24 (0.44–3.49)

10

11

2.82 (1.17–6.82)

    No

1,515

556

1.00

1,515

551

1.00

Ovarian cancer in daughter(s)

    Yes

0

0

0

1

    No

1,525

562

 

1,525

561

 

aAdjusted for age (in categories), pregnancy (never/ever), additional pregnancies (linear) and duration of oral contraceptive use (linear)

Discussion

Comparison of self-reported cancer diagnoses in first-degree female family members with the diagnoses registered in the Danish Cancer Registry showed that self-reports are not always reliable. In our study, the sensitivity depended on the cancer site; that for all cancer was 0.78–0.90, but the sensitivity was far lower (except for breast cancer) and heterogeneous for site-specific cancers, ranging from 0.29 for cancer of the urinary system reported by controls to 0.94 for breast cancer reported by controls. This pattern is similar to previous findings [1, 4,5, 7,10, 12, 16,29]. We saw no substantial differences between cases and controls in the sensitivity, specificity or kappa of reporting of familial cancer. As the case group in our study comprised ovarian cancer patients, it was interesting to see that the sensitivity of reporting ovarian cancer in the closest family was as low for cases as for controls (0.44 and 0.42, respectively). A tendency to under-report cancer in relatives has been seen among persons with other cancer types in which a family history of the disease is a risk factor. Mitchell et al. [2] and Glanz et al. [8] found that patients with colorectal cancer substantially under-reported colorectal cancer in their relatives. In contrast, Chang et al. [12] found that the sensitivity of reporting familial hematopoietic cancer was significantly higher among persons with lymphoma (0.60) than among controls (0.38). Inaccurate reporting of familial cancer diagnoses could be due to unwillingness to talk about cancer or to report a relative’s cancer history, lack of communication among relatives, misinformation, confusion about cancer sites, especially for cancers in the abdominal cavity, or, especially for skin cancers of the non-melanoma type, lack of knowledge about the cancer type or confusion between benign and malignant lesions.

In contrast to the low sensitivity, the specificity of reporting was generally high for the site-specific cancers. This could be expected, as cancer is a disease that family members would be expected to know about, and especially site-specific cancer is relatively uncommon.

We did not find that case or control status was important for accurate reporting, in accordance with some studies [1, 2], but in contrast to that of Chang et al. [12]. Index women tended to report with greater sensitivity about their daughters and sisters than about their mothers, and the specificity with regard to all cancer was significantly higher for daughters than for mothers and sisters and better for sisters than for mothers. The latter finding is in contrast to that of Garbers et al. [13], who found that women recalled the health history of their mothers better than that of their sisters. The mean age of our cohort at the time of interview was 57.8 years, which may explain why they recalled their sisters’ cancer history better than that of their mothers, as some of the mothers would probably have been dead for years at the time of the interview and might have belonged to a generation in which disease was not discussed. Age at interview was found to be a predictor of both sensitivity and specificity (except for the specificity for ovarian cancer) in that there was an overall tendency (significant for all cancer) to report more accurately with younger age at the time of interview. This finding is consistent with those of previous studies [1, 5, 30]. Longer education decreased the sensitivity but increased the specificity for all cancer and breast cancer, as in other studies [1, 10]. This might be due to a difference in the willingness to talk about cancer in families with higher educational levels. We found a slightly increased trend to greater sensitivity with longer interval since a relative’s cancer diagnosis, which is difficult to explain but is in line with the findings of Chang et al. [12]. The increase in sensitivity was regardless of whether the index person was below 60 years or 60 years or above at the time of interview.

When obtaining a family history for the purpose of genetic counseling for hereditary breast and ovarian cancer, the age at diagnosis of relatives with the cancer of interest is important, as hereditary cases often have an earlier debut than non-hereditary cases (except for ovarian cancer in BRCA2 mutation carriers) [31,32]. Cases and controls did not differ significantly in correct reporting of age at diagnosis of any cancer or site-specific cancers: 79–86% of cases and 77–83% of controls gave ages that were within 4 years of those registered. This finding is in line with that of Parent et al. [30].

Estimation of the association between a family history of ovarian or breast cancer and the risk for ovarian cancer on the basis of self-reported family histories of cancer or of validated data from the Danish Cancer Registry revealed substantial underreporting of ovarian cancer in all female relatives, especially by cases, and the risk estimates were almost twice as high and significant when based on Registry data than when based on reported data. The risk estimates for all cancer and breast cancer did not differ substantially whether based on reported or registered data, in accordance with the high sensitivity and specificity for both cases and controls. In a comparison of odds ratios for non-Hodgkin lymphoma based on self-reported data and Swedish Cancer Register data [12], higher odds ratios were found when the analyses were based on self-reported data, in direct contrast to our findings for ovarian cancer.

The strengths of our study are its size, its completeness with regard to identification of first-degree relatives and the possibility of linkage to the Danish Cancer Registry, which is virtually complete from 1943. The participation rates among cases and controls were reasonably high, although 18% of eligible controls participated in a telephone interview only.

Our results suggest that there is bias in self-reporting of family cancer history, which differs by cancer type. We found that age at interview (all cancer) and length of education (all cancer and breast cancer) significantly affected the sensitivity, and type of relative (all cancer and breast cancer), age at interview (all cancer), and length of education (all cancer) significantly affected the specificity. The results indicate that studies of associations with family cancer history should validate self-reported family cancer diagnoses as carefully as possible.

At the individual level, our results indicate that cancer diagnoses in relatives should be verified when hereditary ovarian cancer is suspected, as there is substantial under-reporting of ovarian cancer in relatives. Regarding breast cancer, our data suggest that verification of this disease, including the age at diagnosis in relatives, is less critical for genetic counseling, as self-reported familial breast cancer has been shown to be quite reliable. In order to ensure accurate information and thereby the best counseling, however, we would still recommend that whenever possible pathology reports should be obtained for all affected relatives, particularly if decisions on prophylactic surgery or surveillance are to be based on family history.

Copyright information

© Springer Science+Business Media B.V. 2008