Association between Benign Thyroid and Endocrine Disorders and Subsequent Risk of Thyroid Cancer among 4.5 Million U.S. Male Veterans

Abstract

Context:

Risk factors for thyroid cancer (TC) in males are poorly understood.

Objectives, Setting, and Participants:

Our aim was to evaluate the relationship between history of benign thyroid and endocrine disorders and risk of TC among 4.5 million male veterans admitted to U.S. Veterans Affairs hospitals between July 1, 1969, and September 30, 1996.

Design:

We conducted a retrospective cohort study based on hospital discharge records with 1053 cases of TC.

Main Outcome Measures:

We estimated relative risks (RR) and computed 95% confidence intervals (CI) for TC using time-dependent Poisson regression models. To evaluate potential ascertainment bias and/or delayed diagnosis of TC, we also analyzed RR by time between diagnosis of benign disorder and TC (<5 or ≥5 yr).

Results:

RR for TC were significantly elevated with many disorders and were often higher less than 5 yr compared with 5 yr or more before TC diagnosis. RR (95% CI) less than 5 yr/at least 5 yr were 67.9 (42.4–108.8)/28.9 (9.2–90.2) for thyroid adenoma, 77.8 (64.5–93.1)/25.9 (17.9–38.0) for nontoxic nodular goiter, 23.9 (13.8–41.3)/12.9 (4.8–34.4) for thyroiditis, 8.8 (6.9–11.3)/6.0 (3.8–9.6) for hypothyroidism, 6.4 (4.4–9.4)/ 2.0 (0.8–4.8) for thyrotoxicosis, and 1.2 (1.0–1.4)/1.1 (0.9–1.5) for diabetes. For some disorders, RR also significantly varied by attained age and race with younger patients and Blacks having higher RR than older patients and Whites.

Conclusions:

We found strong associations for a history of thyroid adenoma, nodular goiter, thyroiditis, or hypothyroidism with TC in males allowing for increased surveillance/delayed diagnosis and evidence that some of these associations are modified by age and race.

Incidence of thyroid cancer has been increasing worldwide and has more than doubled in the past 40 yr reaching in 2008 16.3 and 5.6 per 100,000 person per year in females and males, respectively (1, 2). This increase was initially attributed to improved surveillance (3), but further evaluation of the data suggested a true increase in incidence of thyroid cancer of all sizes, implying that increased diagnostic scrutiny is not the full explanation (4, 5). Although the increase in incidence of thyroid cancer has been dramatic, incidence of thyroid cancer remains lower than the respective incidence of breast, prostate, lung, and other common cancers (2).

Both genetic and environmental factors have been implicated in the etiology of thyroid cancer. Thyroid cancer is consistently more common in females, with a female-to-male ratio in incidence rates of 3:1; however, due to delay in diagnosis and more advanced stage at diagnosis, mortality rates are higher and have been increasing more rapidly in males (2, 6). Exposure to ionizing radiation, especially during childhood, is the best-established risk factor for thyroid cancer (7). Preexisting thyroid disorders, including goiter or benign thyroid adenoma, have been associated with high odds ratios (OR) of thyroid cancer in a large, pooled analysis of 12 case-control studies (8). Several environmental factors, including diet, tobacco, and alcohol intake, have been linked with risk of thyroid cancer albeit less convincingly (9). Family history studies suggest that thyroid cancer might have a greater familial component than other cancers with relative risk (RR) estimates of 3–4 or higher in first-degree relatives, (10), although most genetic determinants of risk remain undiscovered (11).

Epidemiological studies used to investigate thyroid cancer risk factors have included primarily females because the majority of cases occur in females. To improve our understanding of risk factors for thyroid cancer in males, we conducted an evaluation of thyroid cancer risk in relation to preexisting thyroid and other endocrine disorders among 4.5 million adult male military veterans admitted to U.S. Veterans Affairs (VA) hospitals.

Patients and Methods

Study population

We conducted an analysis using data from the Patient Treatment File of the U.S. VA medical system (12, 13). The construction of the VA cohort has been described previously (13). Briefly, the study cohort was identified from the VA database of inpatient admission records of 142 U.S. VA hospitals between July 1, 1969, and September 30, 1996. Eligible patients included male White or Black U.S. veterans 18–100 yr old with at least one hospitalization during the study period (14). After excluding patients with any prevalent cancer (8.3%, not including nonmelanoma skin cancer), races other than Black or White (2.3%), females (1.9%), patients younger than 18 or older than 100 yr (<1%), nonveterans (1.6%), and patients who did not survive or were diagnosed with cancer within the first year of first admission (9.3%), the analysis included 4,501,578 patients.

The National Institutes of Health Office of Human Subjects Research granted exemption from institutional review board review and waived informed consent because the study was restricted to existing data with all personal identifiers removed.

Definition of the outcome and preexisting conditions

Thyroid cancer and benign thyroid and endocrine disorders were defined based on the International Classification of Diseases, ICD-8 (until September 30, 1980) or ICD-9-CM (from October 1, 1980 to the end of the study period, September 30, 1996) codes extracted from VA discharge records. Thyroid cancer was defined as ICD-8/ICD-9-CM codes of 193. Benign thyroid disorders were defined as ICD-8/ICD-9-CM codes between 240 and 246. Diagnostic codes for specific thyroid disorders were as follows: nontoxic nodular goiter (ICD-8/ICD-9-CM 241), hypothyroidism (ICD-8/ICD-9-CM 244), simple and unspecified goiter (ICD-8/ICD-9-CM 240), thyrotoxicosis with or without goiter (ICD-8/ICD-9-CM 242), thyroiditis (ICD-8/ICD-9-CM 245), and other thyroid conditions (ICD-8/ICD-9-CM 246). A specific code for thyroid adenoma was available only in ICD-9-CM (226). Benign endocrine disorders (other than thyroid gland disorders) were defined as ICD-8 codes between 250 and 258 or ICD-9-CM codes between 250 and 259, including diabetes, disorders of the pancreas, pituitary, thymus, adrenals, testicles, and parathyroid gland. The most common nonthyroid endocrine disorders included diabetes (ICD-8/ICD-9-CM 250) and disorders of the parathyroid gland (ICD-8/ICD-9-CM 252), which in turn included hyperparathyroidism (252.0), hypoparathyroidism (252.1), disorder of the parathyroid other specified (252.8), and disorder of the parathyroid other and unspecified (252.9). Only benign thyroid or endocrine disorders diagnosed before thyroid cancer or end of follow-up were considered in the analysis. We also defined an additional variable for history of multiple thyroid disorders. This variable was assigned the following levels: no disorder of interest, exactly one disorder, and two or more disorders. When conducting analysis of RR associated with latency of the preexisting condition in patients with multiple thyroid disorders, the latency was determined from the earliest date of the condition of interest to the date of thyroid cancer diagnosis or end of follow-up.

Statistical analysis

We used time-dependent Poisson regression models for grouped survival data to estimate RR and calculate likelihood-based 95% confidence intervals (CI) for the diagnosis of thyroid cancer with different benign thyroid or other endocrine disorders using the AMFIT procedure in Epicure (version 2.0, 1996; HiroSoft International Corp., Seattle, WA). The person-years at risk were counted from 1 yr after the date of first hospital discharge to the date of exit (earliest hospital admission for thyroid cancer, death, or end of study period, September 30, 1996) using the DATAB procedure in Epicure. RR with benign thyroid or other endocrine disorders were adjusted for attained age (18–40, 40–50, 50–60, 60–70, 70–80, 80+ yr), calendar time (U.S. Federal fiscal years 1969–1974, 1975–1979, 1980–1984, 1985–1989, 1990–1996), race (Black or White), time on study (years of follow-up: 2–3, 4–5, 6–10, 10–14, 15+ yr), and number of hospitalizations (1–2, 3–4, 5+) by stratification. All analysis variables except race and number of hospital visits were treated as time-dependent variables, with person-years allocated to disorder and no-disorder categories as appropriate. In addition to the main effect of preexisting condition of interest, we also explored effect modification of RR by latency or time between initial diagnosis of benign disorder and thyroid cancer (<5, 5+ yr), attained age (<50, 50–70, >70 yr), calendar time (before and after January 1, 1980, as a surrogate indicator for the introduction of thyroid ultrasonography, or before and after January 1, 1990, as a surrogate indicator for the introduction of ultrasound-guided fine needle aspiration biopsy) (15), and race (Black, White). Effect modification was tested by comparing nested models with and without an interaction term between the two effects of interest (for example, history of any thyroid disorder and attained age) using the likelihood ratio test. We present effect modification results if these are based on at least three cases per category of the modifying factor. All statistical tests were two sided, with an α level of 0.05.

Results

During 52.7 million person-years of follow-up, 1053 incident cases of thyroid cancer were diagnosed among the 4.5 million U.S. male veterans (Table 1). Median age at study entry for White cases (n = 882) and noncases (n = 3,668,362) was similar (around 53 yr), whereas Black cases (n = 171) were slightly older at study entry compared with Black noncases (n = 832,163; 51.1 vs. 47.7 yr). The mean follow-up time in the cohort as a whole and among noncases irrespective of race was around 11.7 yr. On average, cases in Whites and Blacks were diagnosed 7 and 8 yr after initial hospitalization and had median age at thyroid cancer diagnosis of 60.9 and 61.3 yr, respectively. For both races, cases had a median of four hospitalizations, whereas noncases had a median of three hospitalizations.

Table 1.

Characteristics of the thyroid cancer study cohort of male White and Black U.S. veterans

Variable	White			Black			All
	Case	Noncase	Total	Case	Noncase	Total	Case	Noncase	Total
Patients (n)	882	3,668,362	3,669,244	171	832,163	832,334	1,053	4,500,525	4,501,578
Person-years	6,291	42,758,399	42,764,690	1,376	9,888,117	9,889,493	7,667	52,646,515	52,700,000
Median age at entry (yr)	53.2	53.5	53.5	51.1	47.7	47.7	53	52.5	52.5
Mean follow-up (yr)	7.1	11.7	11.6	8.1	11.9	11.9	7.3	11.7	11.7
Median age at TC diagnosis (yr)	60.9	NA	NA	61.3	NA	NA	60.9	NA	NA
Median no. of hospital visits	4	3	3	4	3	3	4	3	3

NA, Not applicable; TC, thyroid cancer.

The RR for thyroid cancer with history of any thyroid disorder and with specific thyroid disorders were significantly elevated many-fold (Table 2). Very high RR (>40) were observed for thyroid adenoma, nontoxic nodular goiter, and simple/unspecified goiter. Also, there was strong evidence of increase in RR with increasing number of thyroid disorders (P < 0.001). Specifically, individuals with one and two or more thyroid disorders had RR of 30.1 (95% CI = 25.5–35.5) and 48.6 (95% CI = 34.2–69.1) compared with individuals without any thyroid disorder. The RR with history of diabetes, the most common nonthyroid endocrine disorder, was only weakly associated with increased risk of thyroid cancer (1.2, 95% CI = 1.0–1.4). Having a diagnosis of parathyroid disease was associated with an almost 11-fold increase in risk of thyroid cancer; however, this disorder was rare.

Table 2.

RR of thyroid cancer with benign thyroid or endocrine disorders in male White and Black U.S. veterans

Disorder	All			Time to thyroid cancer diagnosis						Pc
				<5 yr			≥5 yr
	n/PYa	RRb	95% CI	n	RR	95% CI	n	RR	95% CI
Any thyroid gland disorder	252/58.7 × 104	23.3	20.1–26.9	199	37.1	31.6–43.7	53	9.5	7.2–12.6	<0.001
Thyroid adenoma	18/9.5 × 103	67.9	42.4–108.8	15	92.8	55.4–155.3	3	28.9	9.3–90.2	<0.001
Nontoxic nodular goiter	145/17.1 × 103	77.8	64.5–93.1	117	141.8	116.3–172.8	28	25.9	17.9–38.0	<0.001
Simple or unspecified goiter	45/44.8 × 103	41.5	30.7–56.1	31	63.4	44.2–90.9	14	23.3	13.7–39.7	<0.001
Thyrotoxicosis with or without goiter	28/18.9 × 104	6.4	4.4–9.4	23	12.4	8.2–18.8	5	2.0	0.8–4.8	<0.001
Hypothyroidism	68/29.7 × 104	8.8	6.9–11.3	49	10.7	8.0–14.4	19	6.0	3.8–9.6	<0.001
Thyroiditis	13/21.3 × 103	23.9	13.8–41.3	9	38.4	19.9–74.2	4	12.9	4.8–34.4	<0.001
Other thyroid disorder	17/25.3 × 103	25.9	16.0–41.9	13	39.8	23.0–69.0	4	12.1	4.5–32.4	<0.001
Multiple thyroid disorders
None	846/52.3 × 106	1.0			1.0			1.0
One	174/31.2 × 104	30.1	25.5–35.5	148	58.5	48.9–70.0	26	7.6	5.2–11.3	<0.001
Two or more	33/31.7 × 103	48.6	34.2–69.1	19	75.4	47.7–119.4	14	31.3	18.4–53.4	0.01
Other endocrine disorders
Diabetes	146/46.1 × 105	1.2	1.0–1.4	75	1.3	1.0–1.6	71	1.1	0.9–1.5	0.11
Disorders of parathyroid glands	17/60.1 × 103	10.7	6.6–17.3	14	19.6	11.6–33.3	3	3.4	1.1–10.6	<0.001

PY, Person-years.

^aNumber of thyroid cancer cases and person-years of follow-up for individuals with preexisting disorder of interest. We chose not to present the number of thyroid cancer cases and person-years of follow-up for individuals without disorder of interest because 1) the number of cases is a difference between 1053, total number of cases, and number of cases with disorder of interest presented in the table and 2) person-years at risk for individuals without specific thyroid/endocrine disorder vary little across all disorders considered and similar to person-years presented for individuals who have no thyroid/endocrine disorders.

^bRR adjusted for attained age, calendar time, race, time on study, and number of hospitalizations by stratification.

^cP likelihood ratio-based test of trend in relative risk by time to thyroid cancer diagnosis.

To evaluate whether increased RR for thyroid cancer observed with diagnosis of thyroid or endocrine disorders might be due to increased surveillance or initial misdiagnosis of thyroid cancer, we analyzed variation in RR by latency or time to diagnosis of thyroid cancer (Table 2). For all disorders of interest except diabetes, the RR exhibited strong, significant variation by latency with RR being much higher within 5 yr of thyroid cancer diagnosis. Therefore, in all further analyses, those individuals whose benign disorders were diagnosed within 5 yr of thyroid cancer were excluded. After this exclusion, significantly increased RR persisted for thyroid adenoma (28.9, 95% CI = 9.2–90.2), nontoxic nodular goiter (25.9, 95% CI = 17.7–38.0), simple/unspecified goiter (23.3, 95% CI = 13.7–39.7), hypothyroidism (6.0, 95% CI = 3.8–9.6), thyroiditis (12.9, 95% CI = 4.8–34.4), other thyroid disorders (12.1, 95% CI = 4.5–32.4), parathyroid disorder (3.4, 95% CI = 1.1–10.6), and number of thyroid disorders. The RR with a history of thyrotoxicosis with or without goiter and diabetes were no longer significant (2.0, 95% CI = 0.8–4.8; and 1.1, 95% CI = 0.9–1.5, respectively). We further evaluated degree of mutual confounding by including history of thyroid adenoma or nontoxic nodular goiter or simple/unspecified goiter into the same model with other thyroid or endocrine disorders under consideration. Although slightly attenuated, the RR with hypothyroidism, thyroiditis, and other thyroid conditions remained significantly elevated, although the RR with parathyroid disorder did not (data not shown). In addition, when hypothyroidism and thyroiditis were analyzed simultaneously, association with thyroiditis was attenuated, but both remained significantly associated with risk of thyroid cancer (5.2, 95% CI = 3.2–9.7; and 5.4, 95% CI = 1.6–17.5, respectively), implying that they are statistically independent.

Analyses of RR according to attained age revealed that younger compared with older patients tended to have significantly higher RR for thyroid cancer with diagnosis of any thyroid disorder including history of any goiter (largely attributed to the effect of nontoxic nodular goiter) and hypothyroidism (Table 3). The RR for endocrine disorders according to three age groups are not shown because these were based on small numbers. Also, there was no significant variation in RR with the conditions of interest by calendar time (data not shown). Table 4 presents the RR with selected thyroid disorders and diabetes according to race. Black patients compared with White patients tended to have significantly higher RR associated with history of any thyroid disorder but not with history of diabetes. The former variation was largely attributed to statistically significant differences in RR with history of any goiter, simple/unspecified goiter, and hypothyroidism among Black and White patients.

Table 3.

RR of thyroid cancer with selected thyroid disorders diagnosed at least 5 yr before thyroid cancer by attained age in male White and Black U.S. veterans

Disorder	Attained age (yr)									Pb
	<50			50–69			≥70
	n	RRa	95% CI	n	RR	95% CI	n	RR	95% CI
Any thyroid gland disorder	14	24.2	14.0–41.9	28	8.1	5.5–12.0	11	6.2	3.3–11.4	0.003
Any goiter	11	24.8	13.4–45.7	18	8.5	5.3–13.6	9	9.5	4.8–18.6	0.04
Nontoxic nodular goiter	7	94.8	44.5–202.0	16	24.9	15.1–41.1	5	13.4	5.5–32.6	0.005
Simple and unspecified goiter	4	63.3	23.5–170.5	5	14.4	6.0–34.9	5	27.4	11.2–66.7	0.11
Hypothyroidism	5	24.8	10.2–60.4	11	5.9	3.3–6.2	3	2.5	0.8–7.9	0.008

n, Number of cases of thyroid cancer with disorder of interest.

^aRR adjusted for calendar time, time on study, race, and number of hospitalizations by stratification.

^bP likelihood ratio-based test of interaction between attained age and disorder of interest.

Table 4.

RR of thyroid cancer with selected thyroid and endocrine disorders diagnosed at least five years prior to thyroid cancer by race in U.S. veterans

Disorder	White			Black			Pb
	n	RRa	95% CI	n	RR	95% CI
Any thyroid gland disorder	38	7.7	5.5–10.8	15	17.5	10.2–30.0	0.02
Any goiter	25	8.8	5.9–13.2	13	19.2	10.8–34.0	0.04
Nontoxic nodular goiter	22	25.5	16.6–39.2	6	27.3	12.1–62.1	>0.50
Simple and unspecified goiter	5	11.2	4.6–26.9	9	64.4	32.8–126.6	0.002
Hypothyroidism	14	4.7	2.8–8.1	5	15.6	6.4–38.2	0.04
Diabetes	56	1.1	0.9–1.5	15	1.2	0.7–2.0	>0.50

n, number of cases of thyroid cancer with disorder of interest; RR, relative risk; 95% CI, 95% confidence interval.

^aRR adjusted for attained age, calendar time, time on study, and number of hospitalizations by stratification.

^bP likelihood ratio-based test of interaction between race and disorder of interest.

Discussion

There have been previous analytic epidemiological studies (Table 5) that evaluated the relationship between preexisting thyroid disorders and risk of thyroid cancer primarily in females (8, 16, 17). Our findings in males support evidence that the strongest and most consistent associations for thyroid cancer are found with history of structural thyroid disorders, particularly nodular goiter and thyroid adenoma. Although the results of a large, pooled analysis of 12 case-control studies that reported high risk with history of goiter in males (38.3, 95% CI = 5.0–291.2) could have been attributed to differential recall bias (8), the results of our study that relied on medically valid data should not. Nevertheless, potentially increased surveillance among patients with preexisting thyroid disorders or initial misdiagnosis of thyroid cancer (a slow growing tumor) is of concern because both could artificially inflate the estimated RR due to increased diagnostic scrutiny or as a result of surgical intervention. If present, such type of bias should be more pronounced in close proximity to thyroid cancer diagnosis and unlikely to occur indefinitely (18). Consistent with this idea, the excess risks for nontoxic nodular goiter and follicular adenoma in our study were greatest within 5 yr of cancer diagnosis, yet strongly elevated RR were present at least 5 yr before cancer diagnosis. Thus, increased case ascertainment or misdiagnosis is unlikely to account for the total risk with history of preexisting nodular goiter or thyroid adenoma.

Table 5.

Summary of epidemiological studies of thyroid cancer in relation to preexisting benign thyroid disorders

Disorder	Study design/source of data
	Retrospective cohort/medical record (n = 1053)a (this study)		Prospective cohort/self-report (n = 40)a (16)		Retrospective cohort/medical record (n = 196)b (17)		Retrospective case-control/self-report (n = 425)a (8)
	RRc	95% CI	HR	95% CI	RR	95% CI	OR	95% CI
Any thyroid gland disorder	9.9	7.2–12.6	4.7	1.6–13.3	NR		NR
Thyroid adenoma	28.9	9.3–90.2	32.8	9.1–117.5	NR		8	9.2–∞
Nodular goiter	25.9	17.7–38.0	9.3	1.3–69.2	3.4	1.8–6.2	38.2	5.0–291.2
Simple goiter	23.3	13.7–39.7	NR		NR		NR
Thyrotoxicosis	2.0	0.8–4.8	2.1	1.1–3.8	0.9	0.3–2.2	3.1	1.0–9.8
Hypothyroidism	6.0	3.8–9.6	1.7	0.02–12.7	0.7	0.4–1.2	1.7	0.3–11.7
Thyroiditis	12.9	4.8–34.4	1.7	0.5–5.3	NR		NR

HR, Hazard ratio; n, number of cases of thyroid cancer; NR, not reported; TC, thyroid cancer; ∞, infinitely large.

^aMales only.

^bMales and females combined.

^cRR of thyroid cancer with selected thyroid disorders diagnosed at least 5 yr before thyroid cancer.

When considering findings of thyroid adenoma in relation to risk of thyroid cancer, several additional points must be kept in mind. Discrimination between thyroid adenomas and benign thyroid nodules in epidemiological studies is challenging because diagnosis of thyroid adenoma is most reliable when based on pathomorphological conclusion and not on self-report or physical exam (19); although a separate diagnostic code for follicular adenoma was available only from 1990, nearly all adenomas in our study were diagnosed after surgery. Intriguingly, recent molecular studies of follicular adenomas have demonstrated that follicular adenomas may contain mutations in RAS and RET genes often found in thyroid cancers and considered to be early events in malignant transformation (20). Together, evidence from epidemiological and molecular studies suggests that an association between thyroid adenoma and thyroid cancer might be causal.

In our study, history of simple/unspecified goiter was associated with significantly increased risk of thyroid cancer comparable in magnitude to risk of nodular goiter or thyroid adenoma. This finding should be interpreted cautiously because of limited resolution of available ICD-8/ICD-9-CM codes. We could not discriminate between true diffuse enlargement of the thyroid gland and diagnosis of goiter without further specification that could have included nodular goiter. In areas of iodine deficiency, diffuse enlargement of the thyroid gland results from insufficient iodine intake, (21) yet in the United States, because of the availability of iodized salt since 1924, this disorder is rare and more likely to occur with Hashimoto thyroiditis or Graves disease (22). Therefore, the intrinsic limitations of our data do not allow additional insight on the important and unsettled issue of association between simple diffuse goiter and thyroid cancer risk, particularly of follicular type (23).

Data are inconsistent concerning the associations between history of preexisting functional thyroid disorders including hyper- and hypothyroidism and risk of thyroid cancer. These disorders are particularly challenging to study because of the lack of standardized definitions across studies, use of different biochemical assays for diagnostic purposes, and potential treatment effects. Although Meinhold et al. (16) found a positive association with self-reported history of thyrotoxicosis in a prospective cohort study, the results from a retrospective cohort study by Iribarren et al. (17) were close to null. In a pooled analysis of 12 case-control studies, the overall RR with history of hyperthyroidism was not significant once controlled for history of goiter or allowing for potential ascertainment bias (8). Similarly, we found that an increased risk with history of thyrotoxicosis in males did not remain significant after adjustment for history of nodular or simple/unspecified goiter or limiting the analyses to cases diagnosed at least 5 yr before thyroid cancer diagnosis. Overall, it appears that available epidemiological data do not support substantial risk of thyroid cancer with history of hyperthyroidism.

Unlike others (8, 16, 17), we found that history of hypothyroidism was associated with significantly increased risk of thyroid cancer even after adjustment for history of nodular or simple/unspecified goiter. TSH is a known growth factor for thyroid nodules and continued TSH stimulation may be associated with initiation and/or progression of thyroid carcinoma (24). In fact, in a retrospective cohort study of over 800 patients undergoing thyroidectomy, higher quartiles of preoperative TSH values predicted pathological identification of thyroid cancer (25). The main cause of hypothyroidism in iodine-sufficient populations, particularly in females, is believed to be autoimmune in origin. There is also emerging evidence from some but not all studies that autoimmune thyroiditis, particularly its most common form, chronic or Hashimoto thyroiditis, might be related to risk of thyroid cancer (26, 27). Hashimoto thyroiditis is a distinct disorder characterized microscopically by the infiltration of lymphocytes and fibrosis and clinically by varying degrees of thyroid dysfunction. Because of the limited information available (ICD-8/ICD-9 codes), we did not know the basis of thyroiditis diagnosis. Although this diagnostic category may have included rare cases of subacute thyroiditis (often caused by viral infection), their contribution to the overall RR observed with history of thyroiditis had to be minimal. More importantly, in our study, association with thyroiditis diagnosed at least 5 yr before thyroid cancer diagnosis was significantly increased and appeared to be independent of hypothyroidism. This finding seems consistent with pathology studies that support an oncogenic potential of autoimmune thyroiditis due to the presence of immunohistochemical markers of papillary thyroid cancer (PTC) (such as cytokeratin 19, p63, and RET/PTC rearrangements) in thyroid tissue affected by the autoimmune process (28).

Because rates for both thyroid cancer and insulin resistance are increasing, the hypothesis has been proposed that these might be related (29). Two recent cohort studies reported a significant association between diabetes and thyroid cancer risk in females but not males (16, 30). By contrast, another study reported an inverse association between blood levels of glucose and thyroid cancer risk in females (31). Our results for males demonstrate that significant association with diabetes was attenuated and lost its significance once individuals with diabetes diagnosed within 5 yr of thyroid cancer were excluded, or association was adjusted for thyroid disorders, implying that increased surveillance of diabetic patients and/or comorbidity with other thyroid disorders likely contributed to the observed association. Although there are biological data suggesting that insulin could act as a growth factor and stimulate thyroid cell proliferation partially through IGF-I-dependent signaling mechanisms (32) resulting in increased thyroid volume and nodularity (33), current epidemiological evidence is inconsistent. To establish whether a relationship between diabetes and thyroid cancer is real, future studies should consider collecting data on history of thyroid disorders and body mass index for relevant analysis. Similar to diabetes, risk associated with parathyroid disorders extending at least 5 yr before cancer diagnosis lacked statistical significance when modeled simultaneously with nodular or simple/unspecified goiter, suggesting a role of increased surveillance/ascertainment.

Data concerning factors that could modify the relationship between preexisting thyroid disorders and thyroid cancer risk are limited due to the number of cases available in individual studies. We explored effect modification by calendar time, attained age, and race. There was no evidence of variation in RR for thyroid cancer with any of the conditions by calendar time. We found stronger associations between history of thyroid disorders, particularly history of any goiter and hypothyroidism, and younger age. The higher RR of thyroid cancer in younger patients with benign thyroid disorders compared with older patients (Table 3) did not seem to be completely accounted for by the lower background rates of thyroid cancer in the younger population, and therefore, other factors such as increased medical scrutiny of younger patients with thyroid diseases might contribute to this finding and warrant further study. Interestingly, similar nonsignificant trends were reported in a pooled analysis of 12 case-control studies, where OR with hypothyroidism, hyperthyroidism, and goiter for PTC diagnosed at younger age were higher than the OR for PTC diagnosed at older age (8).

Based on Surveillance Epidemiology and End Results (SEER) analyses, it is known that overall incidence of thyroid cancer in the United States is lower in Blacks than Whites (34). Certain thyroid disorders such as hypothyroidism and thyroiditis also tend to be less common in Blacks (35). Although the impact of socioeconomic status could be related to differential ascertainment of thyroid cancer in SEER studies, this seems unlikely in our study where the data came from a single medical system. Indeed, the incidence rates of thyroid cancer in Blacks and Whites in our study were comparable, 1.73 and 2.06 per 100,000 persons per year, respectively. However, the thyroid cancer RR associated with history of any goiter, simple/unspecified goiter, and hypothyroidism were higher in Blacks than Whites. It is not clear why Blacks would be more sensitive to thyroid cancer in the presence of goiter or hypothyroidism; perhaps genetic background or other exposures could play a role.

Several strengths and limitations should be considered in the interpretation of our results. Compared with the pooled analysis (8), our study had more than twice as many cases of thyroid cancer in males. This permitted us to explore variation in RR by several factors including race and age. Our patients had access to standardized medical care. The potential for confounding by socioeconomic status was further limited because patients within the VA system typically have lower socioeconomic status (36), and previous VA studies found similar healthcare use rates and outcomes for Whites and Blacks (37, 38). Clinical data were obtained from medical records and thus were not subject to recall bias. The follow-up time of our study was reasonable (average of 11.7 yr, maximum 27 yr).

However, this study included only hospitalized male U.S. veterans limiting generalizability of our findings. Although many VA patients lack other health insurance (36), some diagnoses of interest may have been missed. Less severe thyroid or endocrine disorders are typically diagnosed in outpatient settings and may be underreported in our study. The overall estimates of prevalence of nontoxic nodular goiter, hypothyroidism, and thyrotoxicosis in our cohort were 0.2, 1.2, and 0.4%, respectively. The prevalence of nodular goiter is on the lower side even when compared with other studies in males based on palpation (39). Although this probably reflects underreporting of nodular goiter in the hospital setting, such bias is likely to be nondifferential, and therefore, the estimates of RR are likely to underestimate the true RR. The estimates of prevalence of functional thyroid disorders are even more challenging to compare across the studies due to the lack of standardized definitions and lack of clinical data and laboratory records on diagnostic testing and treatment in our study. Nevertheless, our estimates of hypothyroidism and thyrotoxicosis seem to fall between the respective estimates reported by other studies of overt and subclinical thyroid dysfunction in males (40). In addition, miscoding of discharge diagnoses could have occurred, potentially resulting in misclassification of exposure, but this had to be nondifferential. Also, based on ICD-8/ICD-9-CM codes, we were unable to discriminate between different histological types of thyroid cancer or diagnoses of interest beyond three-digit resolution. Although we did not have data on potential confounders including occupation, radiation exposure, alcohol and tobacco intake, physical activity, diet, or body mass index, these appear unlikely to explain the magnitude of RR observed. Because of increased medical surveillance or initial misdiagnosis of thyroid cancer among patients with benign thyroid or endocrine disorders, related bias is of concern but difficult to evaluate. Allowing for this bias within 5 yr of thyroid cancer diagnosis, the RR remained significantly elevated beyond this time, suggesting that increased surveillance/misdiagnosis could not be the sole explanation for the observed associations. Some of the effect modification results are subject to cautious interpretation due to the limited number of cases per category of the modifying factor.

In summary, our findings provide strong evidence linking thyroid disorders including thyroid adenoma, nodular goiter, hypothyroidism, and thyroiditis with an increased risk of thyroid cancer in males and, furthermore, suggest that some associations are modified by age and race. To clarify these relationships and their underlying mechanisms, pooling data on benign thyroid disorders and biological specimens from large prospective epidemiological studies of thyroid cancer in the context of cohort consortia appears a promising avenue for future research.