Sage Journals: Discover world-class research

Abstract

Objectives

Mammography screening programme sensitivity is evaluated by comparing the interval cancer rate (ICR) with the expected breast cancer incidence without screening, ie. the proportional interval cancer rate (PICR). The PICR is usually found by extrapolating pre-screening incidence rates, whereas ICR is calculated from data available in the screening programmes. As there is no consensus regarding estimation of background incidence, we seek to validate the ICR measure against the PICR.

Methods

Screening data from the three mammography screening programmes of Stockholm, Copenhagen, and Funen in the period 1989-2011 provided data to calculate the ICR. The most commonly described methods of extrapolating pre-screening incidence rates to calculate the PICR were illustrated and PICRs were calculated by year and programme using these different methods and compared with the ICRs.

Results

PICRs varied greatly, reaching a difference of 32–34% in Stockholm, 79% in Copenhagen, and 100–106% in Funen between the highest and the lowest value, depending on which method was applied. PICRs exhibited large variations yearly and from programme to programme. ICRs did not vary to the same extent, ranging on average from 0.100 to 0.136 in the first 12-months and between 0.201 and 0.225 in the last 12-months of the two-year period after a negative screen across the three programmes.

Conclusion

The value of the PICR is hugely influenced by which method is applied, whereas the ICR is calculated purely on data available within programmes. We find that the PICR, the establishing indicator for sensitivity, could preferably be replaced by the ICR.

Keywords

Mammography screening programme evaluation sensitivity interval cancer

Introduction

The purpose of screening mammography is to reduce breast cancer mortality. In order to achieve this, breast cancer cases have to be detected early. In this process there are, however, inevitable side-effects; these include false-negative and false-positive screening tests. A high number of false-negative tests indicates low sensitivity, and a high number of false-positive tests indicate low specificity. It is important to balance sensitivity and specificity within a screening programme, and quality assurance guidelines are used to monitor this balance.

The sensitivity of a screening programme is:

\frac{{numberofbreastcancersinthe invitedpopulationwhichweredeclared positivebythescreeningprogramme}}{numberofbreastcancersintheinvitedpopulation}

which equals

1 - \frac{{numberofbreastcancersintheinvited populationwhichwerenotdeclared positivebythescreeningprogramme}}{numberofbreastcancersintheinvitedpopulation} .

Proportionate Interval Cancer Rate (PICR)

‘Number of breast cancers in the invited population which were not declared positive by the screening programme’ is usually estimated by the number of interval cancers. ‘Number of breast cancers in the invited population is not known, but can be estimated by ‘number of breast cancers expected in the absence of screening’. One of the widely used quality assurance measures to monitor the sensitivity of a screening programme is therefore the proportionate interval cancer

(PICR) = 1 - sensitivity = \frac{numberofintervalcancers}{{breastcancerincidence expectedintheabsenceofscreening}},

where number of interval cancers = breast cancers in screen negative women, including women with a false positive result, clinically presenting before the next scheduled screen, see for example the European Guidelines for Quality Assurance in Breast Cancer Screening and Diagnosis.¹ In the randomized controlled trials the incidence rate in absence of screening was known from the control arm of the trials.² In service screening, no control arm is available, but at the start of service screening it was reasonable to estimate the background rate by the incidence rate prior to start of screening.^3–8

However, with a longer time span since start of service screening, the breast cancer incidence prior to the start of service screening, the breast cancer incidence prior to the start of screening becomes an increasingly inaccurate estimate of the expected incidence in the absence of screening. To remedy this, extrapolations have been made and used as estimates of the background rates,^9–11 but over time the actual development in breast cancer incidence can easily divert from an incidence predicted from past observations.¹² This uncertainty is also reflected in several of the service screening evaluation papers, where the authors have expressed reservation about the prediction methodology.^10,11

Interval Cancer Ratio (ICR)

Another way to estimate 1-sensitivity is the number of interval cancers divided by the sum of the screen detected and interval cancers. We have previously shown that the ICR correlates well with the PICR in data from early rounds of screening programmes.¹³ The ICR measure has the advantage of being based solely on data available in most service screening programmes.

The purpose of this study was to present the PICR and the ICR for the full available screening period of three organized screening programmes in order to evaluate the long-term association between these two indicators of sensitivity and the differences in various means of calculating the PICR. We used data from the screening mammography programmes in Stockholm (Sweden), and in Copenhagen and Funen (Denmark).

Methods

Data sources

The Stockholm population-based screening programme was initiated in August 1989, and by March 1992 all women aged 50–69 had been invited to the first of the biennial rounds. Approximately 25% of the target population had participated in a randomized trial between 1981 and 1989.¹⁴ The Stockholm programme was later expanded to include women aged 40–49 and 70–74, but these age groups were not included in the present study. The Stockholm programme maintains a register including identification of all invited women with dates of invitation, dates of screening, and dates and outcome of all follow-up procedures. Screened women were followed-up weekly for deaths and emigrations out of Stockholm, and for interval cancers in the Swedish Cancer Register. Interval cancers were defined only in screen negative women, not including false-positives. From Stockholm we included data from 1 August 1989 to 31 December 2011.

The Copenhagen municipality population-based screening programme was initiated in April 1991, and by March 1993 all women aged 50–69 had been invited to the first of the biennial rounds.⁷ From the 1^st to 7^th invitation round, age was defined by age at start of the invitation round; from the 8^th round onwards, current age was used as the basis for invitations. The Funen county programme started in November 1993, and by December 1995 all women aged 50–69 had been invited.⁸ The Funen programme used current age as the basis for invitation throughout the programme. The activity in the two programmes, including interval cancers, deaths, and emigrations, has been collected in a Mammography Screening Data Base, including data from the programmes’ administrative systems, the Danish Civil Registration System, as well as a number of other registers.¹⁵ The definition of a primary, incident breast cancer varies slightly across Danish registers. To obtain consistent data therefore, for invasive breast cancer we used only data from the Danish Cancer Registry, and for ductal carcinoma in situ (DCIS) data only from the Danish Breast Cancer Cooperative Group. From Copenhagen we included data from 2 April 1991- 31 December 2007, and interval cancers until 30 November 2009. From Funen, we included data from 2 November 1993- 31 December 2007, and interval cancers until 30 November 2009.

Population-based breast cancer incidence rates for women aged 50-69 were retrieved for Stockholm for 1964–2011 from the Regional Cancer Registry in the municipality of Stockholm and the island of Gotland, for Copenhagen for 1966–2007, and for Funen for 1968–2007 from Nordcan.¹⁶

Definitions

Breast cancer: For screen-detected and interval cancers we included incident, invasive breast cancer (ICD-10 code C50) and DCIS (ICD-10 code D05). Population-based breast cancer incidence rates included invasive breast cancer only.

Screen detected cancers: Breast cancer (invasive or DCIS) found at a screening examination.

Negative screening outcome: A negative screen or a negative assessment /surgery after a positive screen.

Invitation round: Defined administratively by an interval of dates of approximately two years.

Interval cancer: Breast cancer diagnosed after a negative screening outcome within two years of the date of the negative outcome, although excluding breast cancer detection at next screen if this took place earlier.

Interval cancer rate: Number of interval cancers divided by number of women with negative screen outcome within 0–11 months and 12–23 months, respectively, after the screening date.

Background incidence rate: We used pre-screening breast cancer incidence rates for women aged 50–69 as the basis for the various extrapolation methods to estimate the expected breast cancer incidence rate in absence of screening.

Proportionate interval cancer rate (PICR): Interval cancer rate divided by background incidence rate. Four different methods were used for estimating the background incidence rate.

First, linear extrapolation based on the breast cancer incidence from the 5 years prior to the start of screening. Second, the mean of these predicted rates over the entire prediction period. Third, linear extrapolation based on the breast cancer incidence from the 25 years prior to the start of screening. Fourth, the breast cancer incidence rate during the year prior to the start of screening.

Interval cancer ratio (ICR): Interval cancer as a proportion of screen detected and interval cancers.

ICR = \frac{intervalcancers}{intervalcancers + screendetectedcancers}

Analysis

For each programme and year of screening, we tabulated the number of women with negative screens, number of screen detected breast cancers, number of interval cancers, ICR, and the four different estimates of PICR. For each programme and year of screening, graphical comparisons were made between the ICR and the different PICRs.

For the Stockholm data, the analysis was undertaken with STATA, and for the Danish data SAS version 9.2. This study was entirely based on data from registers. No contact has been made with patients, relatives, or their practicing doctors. In Sweden, the study was carried out in accordance with the Swedish Data Protection Law. In Denmark, the study is registered with the Danish Data Inspection Board (number 2008-41-2191).

Results

In Stockholm, the slope of the incidence curve in the last 5 years prior to the start of screening was very similar to the slope of the incidence curve during the entire 25-year period before the start of screening, making the background estimates from these two methods very similar (see Figure 1). A different pattern was seen in Copenhagen and Funen, where the slopes of the incidence curves from the 5 years prior to the start of screening were much steeper than the slopes for the incidence curves from the 25-year period before the start of screening. The mean of the 5-year linear extrapolation initially over-estimated and subsequently under-estimated the background incidence as compared with the 5-year estimate alone. In Stockholm and Copenhagen, the estimate based on the incidence during the year prior to the start of screening was lower than the other estimates almost all the time, while in Funen this was the case only at the end of the study period. In Stockholm, the population-based breast cancer incidence rate for women aged 50–69 after the start of the screening programme was higher than any of the predicted rates, and in Copenhagen and Funen it was higher than two of the four predicted rates. These differences reflect the prevalence peak as women enter screening and the artificial aging during screening ¹⁷, and they should not be mistaken for overdiagnosis, which can be studied only in the birth cohort perspective.¹⁸

Figure 1.

Breast cancer incidence per 100,000 screened women aged 50–69 per year and using four different methods of extrapolating the pre-screening breast cancer incidence for the period 1964–2011 for women living in Stockholm, Copenhagen and Funen.

In all three programmes, the numbers of women with various screening outcomes were lower for the first than for subsequent years, due to slow start-up and/or start later than 1 January (see Tables 1, 2 and 3).

Table 1.

Number of women with normal screening results, number of screen detected breast cancer cases, number of interval cancers, interval cancer rate per 10,000 negative screens, proportionate interval cancer rate (PICR) by various models for the interval 0–11 months & 12–23 months after last negative screen, and interval cancer rate (ICR) by screening year in screening programme in Stockholm.

			0–11 months							12–23 months
Screening Year	Number of women with normal screens	Number of screen detected cancers	Number of interval cancers	Interval cancer rate per 10,000 neg. screens	ICR	PICR	PICR	PICR	PICR	Number of interval cancers	Interval cancer rate per 10,000 neg. screens	ICR	PICR	PICR	PICR	PICR
Screening Year	Number of women with normal screens	Number of screen detected cancers	Number of interval cancers	Interval cancer rate per 10,000 neg. screens	ICR	Extra-polation of 5-year incidence prior to screening start	Extra- pollution of 25-year incidence prior to screening start	Mean of 5-year extra-polation	Incidence rate at end of last year prior to screening start	Number of interval cancers	Interval cancer rate per 10,000 neg. screens	ICR	Extra-polation of 5-year incidence prior to screening start	Extra-polation of 25-year incidence prior to screening start	Mean of 5-year extra-polation	Incidence rate at end of last year prior to screening start
1989	22828	113	15	6.57	0.117	0.31	0.30	0.27	0.32	24	10.51	0.175	0.49	0.48	0.43	0.51
1990	55703	364	31	5.57	0.078	0.26	0.25	0.23	0.27	60	10.77	0.142	0.49	0.49	0.44	0.52
1991	57630	324	45	7.81	0.122	0.35	0.35	0.32	0.38	62	10.76	0.161	0.49	0.49	0.44	0.52
1992	55494	221	32	5.77	0.126	0.26	0.26	0.23	0.28	60	10.81	0.214	0.48	0.48	0.44	0.52
1993	57433	256	38	6.62	0.129	0.29	0.29	0.27	0.32	66	11.49	0.205	0.51	0.51	0.47	0.55
1994	60450	264	50	8.27	0.159	0.36	0.36	0.33	0.40	87	14.39	0.248	0.63	0.63	0.58	0.69
1995	58344	235	45	7.71	0.161	0.33	0.33	0.31	0.37	83	14.23	0.261	0.61	0.61	0.58	0.68
1996	62085	264	56	9.02	0.175	0.38	0.39	0.37	0.43	84	13.53	0.241	0.58	0.58	0.55	0.65
1997	64409	279	58	9.00	0.172	0.38	0.38	0.36	0.43	97	15.06	0.258	0.63	0.64	0.61	0.73
1998	68008	303	62	9.12	0.170	0.38	0.38	0.37	0.44	103	15.15	0.254	0.63	0.63	0.61	0.73
1999	62448	318	62	9.93	0.163	0.41	0.41	0.40	0.48	110	17.61	0.257	0.72	0.73	0.71	0.85
2000	59286	322	61	10.29	0.159	0.42	0.42	0.42	0.50	95	16.02	0.228	0.65	0.66	0.65	0.77
2001	65441	382	64	9.78	0.143	0.39	0.40	0.40	0.47	93	14.21	0.196	0.57	0.58	0.58	0.68
2002	67484	391	67	9.93	0.146	0.39	0.40	0.40	0.48	99	14.67	0.202	0.58	0.59	0.59	0.71
2003	61311	350	59	9.62	0.144	0.38	0.38	0.39	0.46	92	15.01	0.208	0.59	0.60	0.61	0.72
2004	71122	410	47	6.61	0.103	0.26	0.26	0.27	0.32	90	12.65	0.180	0.49	0.50	0.51	0.61
2005	70959	454	68	9.58	0.130	0.37	0.38	0.39	0.46	108	15.22	0.192	0.58	0.60	0.62	0.73
2006	70051	421	60	8.57	0.125	0.32	0.33	0.35	0.41	98	13.99	0.189	0.53	0.54	0.57	0.67
2007	69967	433	64	9.15	0.129	0.34	0.35	0.37	0.44	100	14.29	0.188	0.53	0.55	0.58	0.69
2008	75503	429	54	7.15	0.112	0.26	0.27	0.29	0.34	112	14.83	0.207	0.55	0.56	0.60	0.71
2009	80162	426	65	8.11	0.132	0.30	0.31	0.33	0.39	118	14.72	0.217	0.54	0.56	0.60	0.71
2010	74570	457	58	7.78	0.113	0.28	0.29	0.31	0.37	103	13.81	0.184	0.50	0.52	0.56	0.67
2011	78339	430	61	7.79	0.124	0.28	0.29	0.32	0.37	109	13.91	0.202	0.50	0.52	0.56	0.67

Table 2.

			0–11 months							12–23 months
Screening Year	Number of women with normal screens	Number of screen detected cancers	Number of interval cancers	Interval cancer rate per 10,000 neg. screens	ICR	PICR	PICR	PICR	PICR	Number of interval cancers	Interval cancer rate per 10,000 neg. screens	ICR	PICR	PICR	PICR	PICR
Screening Year	Number of women with normal screens	Number of screen detected cancers	Number of interval cancers	Interval cancer rate per 10,000 neg. screens	ICR	Extra-polation of 5-year incidence prior to screening start	Extra- pollution of 25-year incidence prior to screening start	Mean of 5-year extra-polation	Incidence rate at end of last year prior to screening start	Number of interval cancers	Interval cancer rate per 10,000 neg. screens	ICR	Extra-polation of 5-year incidence prior to screening start	Extra-polation of 25-year incidence prior to screening start	Mean of 5-year extra-polation	Incidence rate at end of last year prior to screening start
1991	10570	103	7	6.62	0.064	0.23	0.26	0.17	0.25	13	12.30	0.112	0.43	0.48	0.32	0.46
1992	19986	190	7	3.50	0.036	0.12	0.13	0.09	0.13	22	11.01	0.104	0.37	0.42	0.29	0.42
1993	17923	126	6	3.35	0.045	0.11	0.13	0.09	0.13	24	13.39	0.160	0.43	0.51	0.35	0.50
1994	13446	67	7	5.21	0.095	0.16	0.19	0.13	0.20	27	20.08	0.287	0.62	0.75	0.52	0.76
1995	14916	88	9	6.03	0.093	0.18	0.22	0.16	0.23	17	11.40	0.162	0.34	0.42	0.30	0.43
1996	16062	93	12	7.47	0.114	0.21	0.27	0.19	0.28	26	16.19	0.218	0.46	0.59	0.42	0.61
1997	11772	58	6	5.10	0.094	0.14	0.18	0.13	0.19	17	14.44	0.227	0.40	0.52	0.37	0.54
1998	15009	83	11	7.33	0.117	0.20	0.26	0.19	0.28	25	16.66	0.231	0.45	0.59	0.43	0.63
1999	13361	53	15	11.23	0.221	0.29	0.39	0.29	0.42	26	19.46	0.329	0.50	0.68	0.50	0.73
2000	13907	80	14	10.07	0.149	0.25	0.35	0.26	0.38	18	12.94	0.184	0.33	0.45	0.34	0.49
2001	16473	98	7	4.25	0.067	0.10	0.14	0.11	0.16	25	15.18	0.203	0.37	0.52	0.39	0.57
2002	14149	90	9	6.36	0.091	0.15	0.21	0.16	0.24	25	17.67	0.217	0.42	0.59	0.46	0.67
2003	10641	64	6	5.64	0.086	0.13	0.19	0.15	0.21	17	15.98	0.210	0.37	0.53	0.41	0.60
2004	9813	60	9	9.17	0.130	0.20	0.30	0.24	0.35	15	15.29	0.200	0.34	0.50	0.40	0.58
2005	16452	139	12	7.29	0.079	0.16	0.24	0.19	0.28	32	19.45	0.187	0.42	0.63	0.50	0.73
2006	16611	106	13	7.83	0.109	0.17	0.25	0.20	0.30	30	18.06	0.221	0.38	0.58	0.47	0.68
2007	18713	122	17	9.08	0.122	0.19	0.29	0.24	0.34	25	13.36	0.170	0.28	0.42	0.35	0.50

Table 3.

			0–11 months							12–23 months
Screening Year	Number of women with normal screens	Number of screen detected cancers	Number of interval cancers	Interval cancer rate per 10,000 neg. screens	ICR	PICR	PICR	PICR	PICR	Number of interval cancers	Interval cancer rate per 10,000 neg. screens	ICR	PICR	PICR	PICR	PICR
Screening Year	Number of women with normal screens	Number of screen detected cancers	Number of interval cancers	Interval cancer rate per 10,000 neg. screens	ICR	Extra-polation of 5-year incidence prior to screening start	Extra- pollutionof 25-year incidence prior to screening start	Mean of 5-year extra-polation	Incidence rate at end of last year prior to screening start	Number of interval cancers	Interval cancer rate per 10,000 neg. screens	ICR	Extra-polation of 5-year incidence prior to screening start	Extra-polation of 25-year incidence prior to screening start	Mean of 5-year extra-polation	Incidence rate at end of last year prior to screening start
1994	17776	160	13	7.31	0.075	0.23	0.29	0.16	0.26	28	15.75	0.149	0.49	0.63	0.35	0.55
1995	22020	227	12	5.45	0.050	0.16	0.21	0.12	0.19	31	14.08	0.120	0.41	0.55	0.32	0.50
1996	21744	100	26	11.96	0.206	0.33	0.46	0.27	0.42	45	20.70	0.310	0.57	0.80	0.47	0.73
1997	22073	136	19	8.61	0.123	0.22	0.33	0.19	0.30	35	15.86	0.205	0.41	0.61	0.36	0.56
1998	22425	111	26	11.61	0.190	0.29	0.44	0.26	0.41	38	16.96	0.255	0.42	0.64	0.38	0.60
1999	22223	105	23	10.35	0.180	0.24	0.39	0.23	0.36	53	23.85	0.335	0.56	0.89	0.54	0.84
2000	23038	128	18	7.81	0.123	0.18	0.29	0.18	0.27	43	18.66	0.251	0.42	0.69	0.42	0.66
2001	22712	147	17	7.49	0.104	0.16	0.27	0.17	0.26	50	22.01	0.254	0.47	0.80	0.50	0.77
2002	22634	132	20	8.84	0.132	0.18	0.32	0.20	0.31	36	15.91	0.214	0.33	0.57	0.36	0.56
2003	24416	125	18	7.37	0.126	0.15	0.26	0.17	0.26	37	15.15	0.228	0.30	0.54	0.34	0.53
2004	24391	146	10	4.10	0.064	0.08	0.14	0.09	0.14	44	18.04	0.232	0.34	0.64	0.41	0.63
2005	24055	139	15	6.24	0.097	0.11	0.22	0.14	0.22	40	16.63	0.223	0.31	0.58	0.37	0.59
2006	21760	142	17	7.81	0.107	0.14	0.27	0.18	0.27	32	15.17	0.184	0.27	0.52	0.34	0.53
2007	25079	220	23	9.17	0.095	0.16	0.31	0.21	0.32	50	19.94	0.185	0.34	0.68	0.45	0.70

In Stockholm the interval cancer rate (per 10,000 normal screenings) varied from a low of 5.57 after screening in 1990 to 10.29 after screening in 2000 for the first 12-months (0–11 months), and from 10.51 after screening in 1989 to 17.61 after screening in 1999 for the last 12-months (12–23 months) (Table 1). In Copenhagen, the interval cancer rates varied more, ranging from a low of 3.35 after screening in 1993 to a high of 11.23 after screening in 1999 for the first 12-months, and ranging from 11.01 after screening in 1992 to 20.08 after screening in 1994 for the last 12-months (Table 2). In Funen, the interval cancer rate varied from 4.10 after screening in 2004 to 11.96 after screening in 1998 for the first 12 months, and from 14.08 after screening in 1995 to 23.85 after screening in 1999 for the last 12 months (Table 3).

The ICR in Stockholm was lowest in 1990 with a value of 0.078, and highest at 0.175 in 1996 for the first 12-months and varied from a low of 0.142 in 1990 to a high of 0.261 in 1995 for the last 12-months (Table 1). In Copenhagen, the ICR ranged from 0.036 to 0.221 in 1992 and 1999, respectively, during the first 12-months, and from 0.104 to 0.329 also in 1992 and 1999, respectively, for the last 12-months (Table 2). In Funen, in the first 12-months the ICR ranged from 0.050 to 0.206 in 1995 and 1996, respectively. In the last 12-months the ICR ranged from 0.120 in 1995 to 0.335 in 1999 (Table 3). The ICRs were fairly similar across programmes, in 2007 ranging from 0.095 to 0.129 during the first 12-months, and between 0.170 and 0.185 during the last 12 months.

Depending on the method for estimating the background rate, the PICRs in 2011 ranged from 0.28 to 0.37 in Stockholm in the first 12-months, and from 0.50 to 0.67 in Stockholm in the last 12-months, respectively (Table 1). The differences between the measures reached 32–34%. In Copenhagen in 2007, the PICRs ranged from 0.19 to 0.34 in the first 12 months, and from 0.28 to 0.50 in the last 12 months (Table 2), thus reaching a difference of 79% for both time intervals. In Funen in 2007, the PICRs varied from 0.16 to 0.32 in the first 12-months, and from 0.34 to 0.70 in the last 12-months (Table 3), thus reaching a two-fold difference in both time intervals. The PICRs furthermore varied considerably across programmes; in 2007 the highest average was 66–67% above the lowest average for both time intervals.

Figure 2 depicts the ICR and the four different PICRs. The PICRs showed that the differences between the four measures increased with time since the start of the screening programmes. In Stockholm and Copenhagen, the highest PICRs were based on the incidence in the last year prior to the start of screening, while in Funen little difference was seen between this measure and the PICR predicted based on the last 25 years. As a consequence of the close relationship between the 5-year and 25-year estimates in Stockholm, the PICRs generated from these two extrapolations were fairly similar. In all three programmes, the PICRs based on the 5-year extrapolation and based on the means thereof were fairly similar, with the expected crossing over in the middle of the period.

Figure 2.

Comparison of ICR and four different means of calculating the PICR per screening year and by 12-month interval in Stockholm, Copenhagen and Funen.

Discussion

The PICR is the recommended indicator of sensitivity in mammography screening. Our study shows that the PICRs could vary up to more than two-fold, depending on how the incidence expected in the absence of screening was estimated. This result questions the usefulness of PICR as a quality indicator.

The smallest differences between the calculation methods were found in the Stockholm programme, where the slope of the incidence rates over the 5-year period was very similar to the slope of the incidence rates over the 25-year period prior to the start of screening. In spite of this, some 20 years after the start of the screening programme there was a 32–34% difference between the highest and lowest PICRs. In the Danish programmes, where the 5-year incidence rates differed from the 25–year incidence rates, the differences in PICRs were much larger. These differences across the three programmes indicate that none of the four extrapolation methods can be recommended universally.

In the European Guidelines, it is stated that an acceptable level for PICR is <= 30% for the first 12-months and <= 50% for the second 12-months after the negative screening outcome.¹ Table 2 shows that by using different choices for estimation of the expected background incidence, all three programmes can be evaluated as reaching or not reaching an acceptable level for PICR. It therefore seems appealing to substitute ‘acceptable values’ and ‘desirable values’ for PICR with ‘acceptable values’ and ‘desirable values’ for another less fluctuating measure, for example, ICR. As the number of screen detected cases is higher in initial screens than in subsequent screens, ICR will be lower in initial screens than in subsequent screens. ‘Acceptable values’ and ‘desirable values’ for PICR would, therefore, have to distinguish between initial screenings and subsequent screenings.

We did not distinguish between initial screens and subsequent screens in this study. Because ICR will be lower in initial screens than in subsequent screens, ICRs cannot be compared across programmes without taking into account the ratio of initial screens in each programme and screening round.

The populations of Denmark and Sweden, as well as the way in which the mammography screening programmes are set up, are quite similar, and the programmes have resulted in reductions in breast cancer mortality of about 25% 19;20), similar to the mortality reduction found in other screening programmes.²¹ This could indicate that the programme sensitivity was acceptable in all three programmes.

The PICR as the recommended measure of sensitivity relies on the estimated background incidence, combined with the interval cancers from the screening programme. There is, however, no consensus on how to estimate the background incidence, which hinders comparison of data across studies. Bordas et al²² used a log linear extrapolation of the incidence rates in the 19-year period prior to the start of screening to estimate the background incidence rate. Day et al⁹ used a linear extrapolation of the incidence rates in the 13-year period prior to the start of screening. Shorter periods have been used more frequently, for example Wang et al, Pellegrini et al, and Hofvind et al,^23,24,6 all using the incidence rates from the 5 years prior to screening as the basis for their estimation of the background incidence rate. Periods ranging from 3 to 5 years were used by Törnberg et al,⁵ and both Kavanagh et al and Woodman et al³ used a 3-year period. Moss et al¹¹ used a combination of the 8-year period prior to the start of screening with results from the Swedish Two County Study to estimate the background incidence.

Incidence rates in non-screened populations from the same area covered by screening have also been used as the basis for the estimates.^25,26 The differences across studies reflect differences in methods as well as differences in availability of pre-screening incidence data. We have illustrated the impact of the estimation method for the resulting PICRs. It can be argued that the PICR to a larger extent reflects the estimation method rather than the underlying sensitivity. The ICR measure, however, does not have this problem, as it is calculated from numbers directly available in most screening programmes. It would therefore be much easier for programmes to report on the ICR than the PICR. We believe that if the gold standard were to estimate ICR, the reporting of interval cancers/programme sensitivity would be much more widespread, enabling much more comparison between programmes.

Instead, the ICR relies on the number of screen-detected cancers, which is affected by length bias and overdiagnosis. In a review of studies that reported overdiagnosis estimates that were adjusted for breast cancer risk and sojourn time, Puliti et al showed that overdiagnosis ranged between 1–11%.²⁷ This amount of overdiagnosis will, for most programmes, not affect ICR as much as the choice of estimate for the background incidence affects PICR. If the breast cancer incidence is stable over time, and the amount of overdiagnosis is high, then PICR would be better than ICR.

Conclusion

Based on the data presented here from three independent, long-term, population-based mammography screening programmes, we find that the PICR, the establishing indicator for sensitivity, could preferably be replaced by the ICR.

Footnotes

Acknowledgement

This study was financially supported by the Nordic Cancer Union and by the PhD Programme of the University of Copenhagen.

Funding

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

References

Perry

Broeders

Tornberg

Holland

von

. European guidelines for quality assurance in breast cancer screening and diagnosis. Fourth edition–summary document. Ann Oncol 2008 Apr; 19(4): 614–622.

Day

Williams

Khaw

. Breast cancer screening programmes: the development of a monitoring and evaluation system. Br J Cancer 1989 Jun; 59(6): 954–958.

Woodman

Threlfall

Boggis

Prior

. Is the three year breast screening interval too long? Occurrence of interval cancers in NHS breast screening programme's north western region. BMJ 1995 Jan 28; 310(6974): 224–226.

Vettorazzi

Stocco

Chirico

Recanatini

Saccon

Mariotto

. Quality control of mammography screening in the Veneto Region. Evaluation of four programs at a local health unit level–analysis of the frequency and diagnostic pattern of interval cancers. Tumori 2006 Jan; 92(1): 1–5.

Tornberg

Kemetli

Ascunce

Hofvind

Anttila

Seradour

. A pooled analysis of interval cancer rates in six European countries. Eur J Cancer Prev 2010 Mar; 19(2): 87–93.

Hofvind

Geller

Vacek

Thoresen

Skaane

. Using the European guidelines to evaluate the Norwegian Breast Cancer Screening Program. Eur J Epidemiol 2007; 22(7): 447–455.

Lynge

. Mammography screening for breast cancer in Copenhagen April 1991-March 1997. Mammography Screening Evaluation Group. APMIS 1998; 83(Suppl): 1–44.

Njor SH, Olsen AH, Bellstrom T, Dyreborg U, Bak M, Axelsson C, et al. Mammography screening in the county of Fyn. November 1993-December 1999. APMIS Suppl 2003;(110):1–33.

Day

McCann

Camilleri-Ferrante

Britton

Hurst

Cush

. Monitoring interval cancers in breast screening programmes: the east Anglian experience. Quality Assurance Management Group of the East Anglian Breast Screening Programme. J Med Screen 1995; 2(4): 180–185.

10.

Everington

Gilbert

Tyack

Warner

. The Scottish breast screening programme's experience of monitoring interval cancers. J Med Screen 1999; 6(1): 21–27.

11.

Moss

Blanks

. Calculating appropriate target cancer detection rates and expected interval cancer rates for the UK NHS Breast Screening Programme. Interval Cancer Working Group. J Epidemiol Community Health 1998 Feb; 52(2): 111–115.

12.

Hakulinen

Teppo

Saxen

. Do the predictions for cancer incidence come true? Experience from Finland. Cancer 1986 Jun 15; 57(12): 2454–2458.

13.

Andersen SB, Tornberg S, Lynge E, von Euler-Chelpin M, Njor SH. Can the burden of interval cancers be measured by the interval cancer ratio? 2014.

14.

Tornberg

Kemetli

Lynge

Helene

Hofvind

Wang

. Breast cancer incidence and mortality in the Nordic capitals, 1970–1998. Trends related to mammography screening programmes. Acta Oncol 2006; 45(5): 528–535.

15.

von Euler-Chelpin

Lynge

Rebolj

. Register-based studies of cancer screening effects. Scand J Public Health 2011 Jul; 39(7 Suppl): 158–164.

16.

http://www-dep.iarc.fr/NORDCAN/english/frame.asp [Accessed 5th March 2014].

17.

Moller

Weedon-Fekjaer

Hakulinen

Tryggvadottir

Storm

Talback

. The influence of mammographic screening on national trends in breast cancer incidence. Eur J Cancer Prev 2005 Apr; 14(2): 117–128.

18.

Njor SH, Olsen AH, Blichert-Toft M, Schwartz W, Vejborg I, Lynge E. Overdiagnosis in screening mammography in Denmark: population based cohort study. BMJ 2013;346:f1064.

19.

The swedish Organised Service Screening Evaluation Group. Reduction in breast cancer mortality from organized service screening with mammography: 1. Further confirmation with extended data. Cancer Epidemiol Biomarkers Prev 2006 Jan; 15(1): 45–51.

20.

Olsen

Njor

Vejborg

Schwartz

Dalgaard

Jensen

. Breast cancer mortality in Copenhagen after introduction of mammography screening: cohort study. BMJ 2005 Jan 29; 330(7485): 220–220.

21.

Njor

Nyström

Moss

. Breast cancer mortality in mammography screening in Europe: a review of incidence-based mortality studies. J Med Screen 2012; 19(Suppl): 33–41.

22.

Bordas

Jonsson

Nystrom

Lenner

. Interval cancer incidence and episode sensitivity in the Norrbotten Mammography Screening Programme, Sweden. J Med Screen 2009; 16(1): 39–45.

23.

Wang

Bjurstam

Bjorndal

Braaten

Eriksen

Skaane

. Interval cancers in the Norwegian breast cancer screening program: frequency, characteristics and use of HRT. Int J Cancer 2001 Nov; 94(4): 594–598.

24.

Pellegrini

Bernardi

Tuttobene

Fanto

Valentini

. Analysis of proportional incidence and review of interval cancer cases observed within the mammography screening programme in Trento province, Italy. Radiol Med 2011 Dec; 116(8): 1217–1225.

25.

Jacobsen

von Euler-Chelpin

Vejborg

Schwartz

Sala

. Seventeen-years overview of breast cancer inside and outside screening in Denmark. Acta Oncol 2013 Jan; 52(1): 48–56.

26.

Taylor

Supramaniam

Rickard

Estoesta

Moreira

. Interval breast cancers in New South Wales, Australia, and comparisons with trials and other mammographic screening programmes. J Med Screen 2002; 9(1): 20–25.

27.

Puliti

Duffy

Miccinesi

de Koning

Lynge

Zappa

Paci

. EUROSCREEN Working Group. Overdiagnosis in mammographic screening for breast cancer in Europe: a literature review. J Med Screen 2012; 19(Suppl 1): 42–56.

Measuring the burden of interval cancers in long-standing screening mammography programmes

Abstract

Objectives

Methods

Results

Conclusion

Keywords

Introduction

Proportionate Interval Cancer Rate (PICR)

Interval Cancer Ratio (ICR)

Methods

Data sources

Definitions

Analysis

Results

Discussion

Conclusion

Footnotes

Acknowledgement

Funding

References