Population-based cancer registries serve a vital role by providing useful information for directing and monitoring cancer control activities and health policy initiatives [1, 2]. Much time and effort has been spent on standardizing the collection, consolidation, analysis, and reporting of cancer surveillance data to ensure that the data are high quality, complete, and comparable, and therefore suitable to aggregate among cancer registries in the United States [3, 4], Canada [5], and within North America [6]. Increasingly, population-based cancer registry data are being used to compare cancer incidence and survival rates among countries worldwide [7–10].

In the mid-1970s, the Surveillance, Epidemiology, and End Results (SEER) Program [11] developed coding rules that helped differentiate a new primary cancer from a distant metastasis or a recurrent cancer. SEER standardized the coding of multiple primary cancers diagnosed in an individual cancer patient, including cancers that were diagnosed concurrent with or subsequent to the first primary cancer [12]. These rules have been revised over time [13–17] and are used throughout the United States and in the majority of Canadian provinces beginning in 2007 [18]. In addition, the International Association of Cancer Registries (IACR) has developed multiple primary coding rules that are used worldwide [19]. In general, SEER rules are more liberal than IACR rules in allowing the registration of multiple primary cancers, particularly cancers that occur in paired organs, at the same anatomic site and with the same histologic type.

To address issues of comparability, IACR has developed a computer software program that processes data about multiple primary cancers reported for the same cancer patient and produces a file consistent with IACR multiple primary rules [20]. This program has been used to process data from North American cancer registries for inclusion in Cancer Incidence in Five Continents [7] and by the CONCORD Programme, when reporting global cancer survival data [9].

Historically, usage of different multiple primary rules has not hindered comparative analyses of population-based cancer survival because estimates have been based on the first primary cancer (i.e., only primary cancer or first of two or more primary cancers) diagnosed in a patient [8–10, 21]. But recently, Brenner and Hakulinen [22] have cautioned against the exclusion of patients with a prior cancer for comparative analyses of population-based cancer survival. As the authors noted, the proportion of cancer patients with a known prior cancer may depend on the length of operation of the cancer registry: Cancer registries that have been in operation for many years will have more information (i.e., past cancer diagnoses) with which to correctly identify a prior cancer than a cancer registry that has been in operation for fewer years. Because the prognosis of a subsequent cancer often differs from that of a first cancer (i.e., the inclusion of all cancers tends to produce somewhat lower survival estimates than those based solely on first cancers) [23, 24], the exclusion of cancer patients with a known prior cancer may introduce a bias in comparative studies because some cancers will be erroneously designated as first cancers when, in fact, they are subsequent cancers. The potential effect of this bias—if estimates are restricted to first cancers-only—will increase as the number and proportion of multiple primary cancers increases because of improved survival for many leading cancers [3, 8, 21, 25]. Furthermore, the mobility and migration patterns of North American populations are variable [26, 27], and a cancer registry may not have knowledge of a previous cancer if that cancer was diagnosed in a different state, province, or country.

International comparisons of cancer incidence [7] and survival [8, 9] are possible, but differences in the coding rules for multiple primary cancers need to be considered when comparing data. Although several studies have compared survival estimates by using first versus multiple cancers using IACR multiple primary rules [22–24], this study is the first, to our knowledge, to compare survival estimates by using both IACR and SEER multiple primary cancer rules among the same patient population.

Figure 1 shows the percentage of multiple primary cancers (all sites and both sexes combined) by year of diagnosis according to SEER and IACR multiple primary cancer coding rules. In 1995, 14.6 and 13.2 %, respectively, of all cancers were reported as multiple primary according to SEER and IACR rules. Between 2005 and 2006, the percentage of multiple primaries declined slightly before continuing to increase. By 2008, the percentage of multiple primary cancers increased to 18.4 % (+25.4 %) according to SEER rules and to 15.8 % (+20.1 %) according to IACR rules.

Table 1 shows the number of first primary cancer sites, by sex, according to SEER multiple primary cancer rules; the number and percentage of additional cancers by using IACR and SEER rules; and the percentage difference between SEER and IACR rules, for all cancer sites combined and the top 25 cancer sites for cancer patients aged 15 years or older at diagnosis and diagnosed between 1995 and 2008 in the SEER-11 area. During 1995–2008, there were 1,015,564 males and 951,022 females diagnosed with at least one primary cancer. According to SEER rules, 189,933 (15.8 %) and 197,370 (17.2 %) additional primary cancers were diagnosed among males and females, respectively. According to IACR rules, 171,363 (14.4 %) and 160,946 (14.5 %) additional primary cancers were diagnosed among males and females, respectively. For each of the top 25 cancer sites and all cancer sites combined, SEER rules identified more multiple primaries than IACR rules. The largest differences in the number of multiple primary cancers between the two rule sets, in rank order, were as follows: melanoma of the skin (5.8 %), urinary bladder (3.5 %), kidney and renal pelvis (2.9 %), colon and rectum (2.7 %), and oral cavity and pharynx (1.6 %) among males; and breast (5.9 %), melanoma of the skin (3.9 %), urinary bladder (3.4 %), colon and rectum (2.9 %), and oral cavity and pharynx (2.2 %) among females.

Table 2 shows 1-, 3-, 5-, and 10-year age-standardized relative survival (RS%) estimates and 95 % CI for all cancer sites combined and for the top 25 cancer sites among males and females for each data set (First, SEER MP, and IACR MP) and differences for First versus SEER MP and IACR MP, respectively, and between SEER MP and IACR MP data sets. Differences where the point estimate from one data set was not contained within the confidence interval for the comparative data set were noted. Compared with estimates using first primary cancers-only, those that included all first primary cancers matching the selection criteria gave more conservative (i.e., lower) 5-year survival estimates for all sites combined (SEER: RS −1.30 %, IACR: RS −1.29 %) among males and (SEER: RS −0.23 %, IACR: RS −0.55 %) among females, respectively; and for 11 of the top cancers: oral cavity and pharynx (RS −2.26 %, RS −2.20 %), colon and rectum (RS −1.31 %, RS −1.30 %), larynx (RS −1.62 %, RS - 1.59 %), melanoma of the skin (RS −1.05 %, RS −1.12 %), prostate (RS −1.06 %, RS −1.07 %), urinary bladder (RS −2.37 %, RS −2.06 %), kidney and renal pelvis (RS −0.78 % SEER only), thyroid (RS −1.90 %, RS −1.92 %), Hodgkin’s lymphoma (RS −1.28 %, RS - 1.22 %), non-Hodgkin’s lymphoma (RS −0.76 %, RS −0.68 %), and leukemia (RS −0.91 %, RS −0.86 %) among males and oral cavity and pharynx (RS −1.73 %, RS −1.74 %), colon and rectum (RS −0.69 %, RS −0.69 %), larynx (RS −2.72 %, RS −2.72 %), melanoma of the skin (RS −0.89 %, RS −0.88 %), breast (RS −0.76 %, RS −0.67 %), corpus and uterus, NOS (RS −1.40 %, RS −1.40 %), urinary bladder (RS −2.54 %, RS −2.28 %), kidney and renal pelvis (RS −0.89 % SEER only), thyroid (RS −0.78 %, RS −0.77 %), non-Hodgkin’s lymphoma (RS −0.77 %, RS −0.68 %), and leukemia (RS −1.99 %, RS −1.95 %) among females. Five-year cancer survival estimates differed between SEER MP and IACR MP for all sites combined (RS 0.32 %) among females.

This study provided a unique opportunity to evaluate the effect of two widely used multiple primary cancer coding rules, SEER and IACR, on population-based survival estimates using SEER-11 cancer survival data, covering approximately 14 % of the US population.

Until recently, the presence of multiple primary cancers and the rules used to collect and report these cases were generally not considered in survival analyses because population-based survival estimates were based on first primary cancers—either the only cancer diagnosed in a patient or the first of multiple primary cancers diagnosed [8, 9, 21]. However, this practice is no longer advocated because it can result in the exclusion of a relatively large and growing number of cancer patients and can lead to the introduction of biases into survival analyses [22].

In the United States, the percentage of multiple primary cancers by using SEER rules is relatively large and increasing. The percentage was higher in females compared to males (17.2 vs. 15.8 %) reflecting, in large part, the influence of these rules on female breast cancer counts. Between 1995 and 2008, the percentage of multiple primary cancers increased 25.4 % among SEER-11 registries. This increase reflects improved survival for many adult cancers and the fact that cancer patients remain at risk, and in some cases increased risk, of developing subsequent primary cancers [32]. The SEER rules have changed over time and field studies to test the most recent rules began in late 2005 and may account for the slight reduction in the percentage of multiple primary cancers observed between 2005 and 2006. The rules were modified to include longer time requirements between same site primaries for certain sites [17].

This study confirmed the finding from previous studies [23, 24] that survival estimates based on first cancers-only exclude a large, varied, and increasing number of subsequent primary cancers and generally produce less conservative (i.e., higher) 5-year survival estimates than estimates produced by using all primary cancers based on SEER or IACR multiple primary cancer coding rules. Therefore, attempting to exclude subsequent cancers could introduce a systematic bias into survival analyses if the rules for registering multiple primaries are not applied consistently among all registries, or over time within a registry, or if information on previous cancer diagnoses is missing.

Clinicians may intuitively recognize a new primary cancer from that of a distant metastasis, extension, or recurrent cancer. However, it can be difficult to codify the rules for reporting these cancers by using routinely collected surveillance data. The identification of multiple primary cancers may reflect the following: (1) the quality and completeness of the surveillance data being reported to the cancer registry by hospitals, physicians’ offices, and laboratories and (2) the training of hospital and registry staff to code the data and interpret the rules. These coding rules may be particularly challenging for registries that use SEER rules because these rules are more complex and require more detailed information than IACR rules. Cancer registries may lack information on previous cancers if the patient was diagnosed in a different state, province, or country from their current residence. This lack of information could be problematic for cancer registries in the United States because registries operate independently in each of the 50 states, the District of Columbia, and six metropolitan areas [6] with support from one or both of the two federal cancer surveillance programs: the National Cancer Institute’s SEER Program [11] or the Centers for Disease Control and Prevention’s National Program of Cancer Registries (NPCR) [4]. NPCR-funded cancer registries could be particularly impacted because many of them have been fully operational since the mid-1990s or later, and SEER metropolitan area registries, with substate catchment areas.

Furthermore, survival estimates based on first cancers-only do not represent the clinical experience of the entire cancer patient population [23]. For example, the introduction of screening practices, such as the prostate-specific antigen (PSA) test, may advance the diagnosis date of prostate cancer by many years before the cancer would be, if at all, clinically detectable. Early-stage prostate cancers can have high survival rates [32], and a patient may die from a subsequent cancer (e.g., colorectal) rather than their prostate cancer. Including both prostate and colorectal primary cancers in survival analyses would be clinically and epidemiologically relevant.

Although it clearly makes sense to use all primary cancers in population-based survival studies, the question remains: Does the choice of SEER or IACR multiple primary cancer coding rules affect survival estimates? This study did not find any appreciable difference between site-specific survival estimates based on SEER or IACR multiple primary rules. SEER multiple primary coding rules are more liberal than IACR rules in allowing the registration of multiple primary cancers, particularly cancers that occur at the same site including paired organs (breast, kidney) and organs with relatively large surface areas (skin, urinary bladder, colon and rectum, oral cavity and pharynx). However, for both the SEER MP and the IACR MP analyses, cases were selected based on using the first cancer that matched to the selection criteria (first cancer in primary site category). This selection criterion appears to have mitigated the influence of SEER multiple primary rules. The one exception was for all sites combined among females where SEER rules produced marginally higher survival estimates than IACR rules (RS 62.28 vs. 61.96 %, respectively). In preparing the data sets, the IACR rules were applied to data from all diagnosis years, including those prior to the reference years for this study (1995–2008) that resulted in the exclusion of some cases that the SEER rules identified as multiple primary cancers. Hence, the case counts differed for the all sites combined category (Table 1) as did the survival estimates (Table 2).

In conclusion, our results showed that survival estimates based on SEER multiple primary rules, which are used by cancer registries in the United States and in the majority of provincial registries in Canada [18], are comparable to those based on IACR rules, which are used by cancer registries throughout the remaining world, when site-specific analyses include all first primary cancers matching the selection criteria. Therefore, to produce clinically and epidemiologically relevant and less biased cancer survival estimates, researchers may consider including data on all cancers in their analysis. However, the multiple primary rules used to collect and record the cancer case data do not affect survival estimates if all first cancers matching the selection criteria are used to produce site-specific survival estimates.

Population-based cancer survival is an important outcome measure for assessing cancer control and health policy initiatives. Together with information on cancer incidence and death, survival data can be used to evaluate progress in preventing, screening, diagnosing, and treating cancer patients [33]. To facilitate the comparison and interpretation of international data and to align the incidence data more closely with the cases used in the survival analyses, cancer registries in North America may want to consider publishing population-based incidence and survival statistics that are based on IACR multiple primary cancer coding rules.

Survival estimates from this study were calculated for the purpose of comparing SEER and IACR multiple primary cancer coding rules and were not intended to represent the survival experience of cancer patients in the study area. Life expectancy varied among geographic areas in the United States [9, 34], and estimates were not adjusted for state-specific life expectancy because these life tables are not currently available. Life tables were based on the life expectancy of the general population. As such, these tables may overestimate life expectancy for individuals with subsequent cancers.