Risk factors for radiation-induced hypothyroidism

A literature-based meta-analysis

Authors


Abstract

BACKGROUND:

A systematic overview and meta-analysis of studies reporting data on hypothyroidism (HT) after radiation therapy was conducted to identify risk factors for development of HT.

METHODS:

Published studies were identified from the PubMed and Embase databases and by hand-searching published reviews. Studies allowing the extraction of odds ratios (OR) for HT in 1 or more of several candidate clinical risk groups were included. A meta-analysis of the OR for development of HT with or without each of the candidate risk factors was performed. Furthermore, studies allowing the extraction of radiation dose-response data were identified for a meta-analysis of the dose-response curve.

RESULTS:

Female gender (OR = 1.6; 95% confidence interval [CI], 1.3-1.9; P < .00001), surgery involving the thyroid gland (OR = 8.3; 95% CI, 5.7-12.0; P < .00001), or other neck surgery (OR = 1.7; 95% CI, 1.16-2.42; P = .006) were associated with a higher risk of HT. Caucasians were at higher risk of HT than African Americans (OR = 4.8; 95% CI, 2.8-8.5; P < .00001). The data showed association between lymphangiography and HT but with evidence of publication bias. There was a radiation dose-response relation with a 50% risk of HT at a dose of 45 Gy but with considerable variation in the dose response between studies. Chemotherapy and age were not associated with risk of HT in this analysis.

CONCLUSIONS:

Several clinical risk factors for HT were identified. The risk of HT increases with increasing radiation dose, but the specific radiation dose response varies between the studies. The most likely cause of this heterogeneity is differences in follow-up between studies. Cancer 2011;. © 2011 American Cancer Society.

External beam irradiation of the cervical region has been used for decades in the management of head and neck carcinomas, lymphomas, and malignancies of the central nervous system. Even the most modern radiation therapy (RT) will inevitably cause incidental exposure of nontarget tissues and organs. This may lead to functional consequences that are readily apparent to both patients and clinicians. However, radiation damage to the thyroid gland can be less conspicuous.

Several types of thyroid effects have been reported.1 The most common clinical late effect of thyroid gland irradiation in patients exposed to therapeutic doses (30 Gy-70 Gy) to the neck is primary hypothyroidism (HT).2 HT may be clinically apparent, with high thyroid-stimulating hormone (TSH) and low free T4 (fT4), or subclinical, with high TSH and normal fT4, but symptoms may not manifest until marked biochemical HT exists. Symptoms include fatigue, edema, myalgia, dry skin, and depression—symptoms that are unspecific. If untreated, severe HT may cause hypercholesterolaemia and impaired consciousness (myxoedema coma). Fortunately, HT can be treated with levothyroxin, and the treatment is relatively well tolerated.3

Initially, radiation-induced HT (RIHT) was considered a rare complication of treatment; the first case was reported in 1961, after therapy for laryngeal carcinoma.4 The incidence of RIHT depends on multiple factors, most important being the location of the radiation fields and the diagnostic intensity and sensitivity, but the typically reported range is 20% to 30% among patients receiving RT that includes the neck.5 Half of these events will appear within the first 5 years of follow-up, with a peak incidence after 2 to 3 years.5 For comparison, the prevalence of HT in the US is 9.5%, with 0.4% being clinically apparent and the remainder being subclinical HT.6

Evidence suggests a dose-dependent risk of RIHT, but the dose-response relation is not well established.7, 8

Thus, as RIHT is readily treatable and relatively common, with symptoms markedly influencing quality of life, it is important to be able to estimate the risk of RIHT to properly inform the patients of treatment side effects. Also, the importance of recognizing HT during follow-up should be emphasized. A recent study showed that intensity-modulated radiotherapy (IMRT) plans for locally advanced head and neck cancer may increase the radiation dose to the thyroid compared with conventional 3D conformal RT unless appropriate thyroid dose constraints are implemented.9 An improved ability to quantify a dose-response relation will at least in some cases help facilitate a rational optimization of RT risk:benefit.

In this article, we present a systematic review of published studies reporting RIHT risk and perform a meta-analysis of suspected clinical risk factors for this side effect. Furthermore, we derive a dose-response relation by fitting dose-response functions for individual studies followed by a meta-analysis of the parameters defining these relations to derive a synthesized best estimate of the radiation dose response of the thyroid gland.

MATERIALS AND METHODS

Search Strategy and Selection Criteria

English-language studies published after 1990 were identified by searching the PubMed and Embase. The final search was performed on July 8, 2010. The search string in PubMed was ((((hypothyroidism and radiotherapy) NOT radioiodine) NOT iodine[Title]) NOT review[Publication Type]) NOT thyroidectomy[MeSH Terms] NOT (“bone marrow” AND transplant*) NOT TBI. A similar search structure was built in Embase. Additional studies were identified by manual screening of the reference lists of included studies and selected reviews. Studies allowing the extraction of odds ratios (OR) with confidence interval (CI) for clinical or biochemical hypothyroidism (any grade and scale) in 1 or more of the following candidate clinical risk groups were included: use of chemotherapy (yes vs no), race (Caucasian vs African American), sex, surgery (yes vs no), and age (older vs younger). We posed no limit on the actual cutoff between the “young” and “old” age groups in the test for age as a risk factor. Although the exact cutoff is not expected to influence the OR for adults, extrapolation of the results to age groups far from the cutoff in the included studies will not be possible. Furthermore, studies reporting radiation dose-response data for hypothyroidism were included.

We included reports on patients receiving radiotherapy in the region of the thyroid gland alone or as part of a multimodality therapy for any malignancy except cancer of the thyroid gland itself, but we excluded studies with 3 or fewer cases of hypothyroidism. Studies where the patients did not receive radiation to the thyroid gland (cranial irradiation only) were excluded, as hypothyroidism caused by damage to the pituitary gland and hypothalamus cannot be assumed to share risk factors with hypothyroidism caused by radiation damage to the thyroid gland. Studies in which both the hypothalamus-pituitary (HP) chain and the thyroid were irradiated were included only if data on primary HT, caused by damage to the thyroid, could be separated from central HT caused by damage to the HP chain. Studies involving total body irradiation were excluded as the use of low radiation doses and intense medical therapy associated with this treatment mean that the incidence of HT after total body irradiation may not be attributable to RT and may not share risk factors with more conventional RT. We imposed no restrictions on study design.

For the extraction of dose-response data, we used the same inclusion/exclusion criteria as above regarding patients, interventions, outcome, and study design. To be included in the meta-analysis of dose response, studies should allow extraction of the incidence and its standard error versus thyroid dose for at least 2 dose levels. This was required to allow a within-study fit of the dose-response curve. The parameters from the within-study dose-response fits can then be combined in a meta-analysis by synthesizing the steepness and position of the dose-response curve across the studies. The approach is a generalization of the synthesis of OR for development of HT for patients with/without a dichotomous risk factor (eg, surgery) to the case of a continuous risk factor (radiation dose). The advantage of the method is that the principle of synthesizing within-study measures of effect size is retained. The method was described in more detail in a recent study of dose-response data for biochemical control of prostate cancer.10

The steepness of the dose-response curve is quantified by the normalized dose-response gradient, γ50, defined as the absolute increase (in percentage points) in the incidence of RIHT for a 1% relative increase in the dose producing a 50% risk of RIHT (D50).11 Hence, a large value of γ50 indicates a steep dose-response curve where a small change in dose results in a large change in the risk of complications.

Statistical Analysis

OR and CI were extracted from the original studies as follows. If reported, the OR and CI were extracted from a multivariate analysis. Otherwise, the OR and CI were extracted from a univariate analysis or alternatively calculated on the basis of the reported number of hypothyroid patients and the total number of patients in each risk group. If the original studies reported the OR with a P-value, but without CI, and the hypothyroid/total number of patients was unavailable, the standard error of the OR was approximated as follows: First, the z0-value of the normal distribution corresponding to the reported P-value was found from statistical tables. By noting that the logarithm of the OR is normally distributed under the null hypothesis, the standard error of the OR, SOR can be estimated by equation image. The assumption in this approximation is that the P-value reported in the original text is consistent with a test of ln(OR) = 0.

The Mantel-Haenszel method was used for data extracted from the original publications as the number of patients with RIHT/total number of patients. If OR and CI were extracted from more than 1 study, the ln(OR) estimates were pooled and weighted by the inverse variance of the logarithm of the odds ratio.

The results are displayed in forest plots with the studies ordered according to statistical weight; this provides a visual impression of effect size as a function of statistical power. A decrease in effect with increasing power might be suggestive of publication bias, in which case a formal test for trend was performed by a regression analysis of the logarithm of the OR versus the inverse of the sample size as recommended by Peters et al.12 Heterogeneity between studies was tested using both Cochran's Q and I2, where the latter is a measure of the percentage of variation that is attributable to heterogeneity rather than chance. Some authors prefer I2 to Cochran's Q because the latter has been criticized for having a power to detect heterogeneity, which increases with the number of studies included in the analysis.13 Statistical analyses were performed using the Review Manager software version 5.0.24,14 except tests of publication bias, which were performed in SPSS 15.0 (IBM SPSS, Chicago, IL). We used fixed effects modeling unless otherwise noted. A 2-tailed P < .05 was considered statistically significant; a Bonferroni correction was used to adjust for multiple comparisons where relevant.

For the dose-response analysis, the proportion of patients with RIHT relative to the total number of patients was plotted versus mean thyroid dose as reported in the published studies fulfilling the above-mentioned criteria for inclusion in the meta-analysis of dose response. A logistic dose-response function parameterized according to Bentzen and Tucker11 was fitted to data from a single study using the glmfit method in the statistical toolbox of Matlab.15 The values of γ50 and D50 and their standard error where extracted and a pooled analysis of γ50 was made by inverse variance weighting in Review Manager.

RESULTS

In total, 467 unique records were identified from the electronic databases and an additional 16 records were identified by hand search. Of the 483 screened records, 351 were excluded based on the title or abstract, predominantly from 1 of the following causes: case reports, reviews, model systems (not human data), management of RIHT rather than incidence, studies without RT of the thyroid region, or reports on radioactive iodine therapy.

Of the 132 articles selected for full text screening, 98 were excluded. In addition to the reasons mentioned above, exclusion at this second pass was predominantly due to lack of data amenable for analysis or study cohorts overlapping more recent reports.

In total, 33 articles involving 8,138 patients were included in the analysis of clinical risk factors,1, 7-9, 16-44 whereas 4 studies involving 1027 patients were included in the analysis of dose-response data.7, 23, 32, 45 The majority of the included studies used biochemical hypothyroidism as an end point as detailed in Table 1.

Table 1. Summary of Included Studies
StudyDiagnosisNo. of PatientsStudy DesignComparisonsEnd Point Definition
  • TSH indicates thyroid-stimulating hormone; TRH, thyroxine-releasing hormone; H&N, head and neck carcinoma (nonthyroid); HT, hypothyroidism.

  • a

    The cutoff between young and old age group is given in parentheses.

  • b

    Biochemical data were available, but the clinical HT definition was used because TSH were not measured consistently.

  • c

    TSH levels were measured every 6 months, but the specific end point was not defined in the original article.

  • d

    Data on chemotherapy were in the original article but are excluded because the total number of HT patients in the chemotherapy analysis is not in accordance with the reported number of HT patients in the article.

Bethge et al16Hodgkin lymphoma177RetrospectiveChemotherapy, age (38 y)a, sexElevated TSH or elevated response to TRH
Bhandare et al7H&N312RetrospectiveChemotherapy, race, dose effectClinical primary hypothyroidismb
Brusamolino et al17Hodgkin lymphoma120ProspectiveSexUndefined HTc
Bramswig et al18Hodgkin lymphoma600RetrospectiveLymphangiogramElevated TSH
Chin et al19Medulloblastoma & primitive neurodectermal48RetrospectiveChemotherapy, sex, age (8 y)aElevated TSH
Colevas et al20H&N118RetrospectiveSex, age (60 y)a, surgeryElevated TSH
Diaz et al10H&N168RetrospectiveSexElevated TSH
Fein et al21Hodgkin lymphoma104RetrospectiveLymphangiogram, chemotherapy, age (23 y)aElevated TSH
Garcia-Serra et al22H&N504RetrospectiveSurgeryElevated TSH
Grande et al23H&N221RetrospectiveSurgery, dose effectElevated TSH
Hancock et al1Hodgkin lymphoma1787RetrospectiveChemotherapyElevated TSH
Ho et al24H&N147RetrospectiveSex, age (70 y)aElevated TSH
Illes et al25Hodgkin lymphoma151ProspectiveLymphangiogram, sex, age (30 y)aElevated TSH
Kanti et al26H&N187ProspectiveChemotherapyElevated TSH
Khoo et al27Hodgkin lymphoma320RetrospectiveLymphangiogramElevated TSH
Koontz et al28Hodgkin lymphoma111RetrospectiveChemotherapy“HT requiring pharmacologic replacement”
Kumpulainen et al29Larynx72ProspectiveAge (62 y)aElevated TSH
Kuten et al45Hodgkin lymphoma33ProspectiveDose effectElevated TSH
Liening et al30H&N96RetrospectiveSurgeryElevated TSH
Lo Galbo et al8Larynx, hypopharynx156RetrospectiveChemotherapy, sex, age (60 y)a, surgeryElevated TSH
Mercado et al31H&N143RetrospectiveChemotherapy, sex, age (50 y)a, raceElevated TSH
Metzger et al32Hodgkin lymphoma461RetrospectiveChemotherapy, sex, race, dose effectElevated TSH
Moser et al33Hodgkin lymphoma757RetrospectiveSexNot stated
Nair et al34Hodgkin lymphoma35ProspectiveLymphangiogramElevated TSH
Paulino35Medulloblastoma32RetrospectiveChemotherapyPrimary clinical HT (elevated TSH and low thyroxine)
Peerboom et al36Hodgkin lymphoma81Recall of patients for measurementsSex, age (20 y)a,dElevated TSH
Ricardi et al37Medulloblastoma32ProspectiveSexElevated TSH
Tell et al38H&N391ProspectiveSex, surgeryClinical primary HT and overt HT (elevated TSH and low thyroxine/free thyroxine)
Thorp et al39Larynx, hypopharynx28RetrospectiveSurgeryElevated TSH
Turner et al40H&N104RetrospectiveChemotherapy, surgeryElevated TSH
Ulger et al41Nasopharynx85RetrospectiveSexElevated TSH
Weissler42H&N68RetrospectiveChemotherapy, sex, surgeryElevated TSH
Wu et al43Nasopharynx408ProspectiveAge (30 y)a, sex, chemotherapyElevated TSH
Zakotnik et al44H&N114RandomizedChemotherapyElevated TSH

Figure 1 shows a flow chart of the search and study selection. Table 1 presents a short summary of the included studies in terms of disease site, number of patients, study design, and end-point definition of the original study along with a list of the risk factors that could be extracted from the studies.

Figure 1.

A flow chart depicts the study's selection process.

Clinical Risk Factors

The result of the synthesis of clinical risk factors is presented in Figure 2. In total, 8 statistical tests were made, so by applying the Bonferroni correction to the single-test nominal significance level of P = .05, a P < .0063 should be considered significant. The meta-analysis showed that female gender is a significant risk factor for RIHT (OR = 1.57; 95% CI, 1.30-1.88) with no evidence of heterogeneity between studies.

Figure 2.

The clinical risk factors for hypothyroidism are shown: (A) sex; (B) age; (C) chemotherapy; (D) surgery; (E) Caucasian versus African descent; (F) lymphangiography (LAG) performed.

Age does not appear to influence the risk of RIHT. Note that the confidence interval of the pooled estimate is 0.89-1.08. Thus, a potential effect below the detection power of this study is likely to have limited clinical significance. The cutoff between the young and old age group in the included studies varies from 20 to 70 years, except for a single study using 8 years as cutoff.19 Consequently, extrapolation of the result to pediatric patients may not be valid.

Chemotherapy did not affect the risk of RIHT in this analysis. Some heterogeneity exists among studies. The study of Hancock et al1 spanned almost 3 decades (from 1961-1989) with a gradual increase in the use of chemotherapy. The authors therefore admit a possible confounding from improved thyroid hormone testing over the years, which might mistakenly have indicated a detrimental effect of chemotherapy. The study by Koontz et al,28 on the other hand, used a higher dose of RT in the single-modality treatments (mean RT dose, 37.9 Gy) compared with the chemoradiation treatments (mean RT dose, 25.5 Gy). This would tend to confound the data toward chemotherapy protecting against HT. Excluding those 2 studies from the synthesis, we found that the odds ratio for HT with chemoradiation compared with RT only was 1.01 (95% CI, 0.77-1.33; P = .94) and the heterogeneity disappeared (Cochran's Q, P = .44; I2 = 1%; data not shown).

Surgery was categorized according to procedures involving the thyroid gland versus those that did not. In both groups, patients receiving RT and surgery were compared with RT alone. We noted a large and highly significant influence of partial or hemithyroidectomy (OR = 8.28; 95% CI, 5.71-12.02). In comparison, surgery not involving the thyroid gland carried a smaller but still significant risk (OR = 1.68; 95% CI, 1.16-2.42; P = .006). There was some level of heterogeneity in the latter data, suggesting an influence of the surgeon or the selection of patients. The test for difference between the 2 types of surgery was also highly significant (P < .00001).

Three studies7, 31, 32 investigated the influence of Caucasian versus African American descent. The 2 smaller studies are in good agreement with the large study by Metzger et al, and concluded that being Caucasian is a large and highly significant risk factor (OR = 4.84; 95% CI, 2.76-8.49; P < .00001).

Five studies allowed extraction data on the possible influence of lymphangiography (LAG).18, 21, 25, 27, 34 The pooled estimate was significant, but there was some heterogeneity in data. We especially noted that the point estimates of the OR tended to decrease with the weight of the studies. A test for publication bias12 yielded a P-value of .097.

Dose-Response Analysis

Four studies comprising 1027 patients fulfilled the inclusion criteria of the dose-response meta-analysis.7, 23, 32, 45 Two additional studies8, 24 allowed extraction of dose-response data for 2 dose levels, where the low-dose level was combined surgery and RT, whereas the high-dose level was RT only. Those 2 studies were excluded as the analysis above demonstrated that surgery is a significant risk factor. Both of the excluded studies showed a negative dose-response relation, ie, the high-dose arm was less toxic than the low-dose arm. In addition, 2 studies were excluded as the high-dose arm was given as hyperfractionated RT.20, 37 Three studies investigated the influence of radiation dose on RIHT risk but did not provide data amenable for the meta-analysis, most often because the number of patients at risk was omitted from the survival curves, thus precluding estimation of CI and, consequently, data weights;46, 47 all 3 studies demonstrated a positive dose-response relation.

The dose in the lowest-dose bin of 1 of the included studies was estimated to 30 Gy to 40 Gy in accordance with the width of the other dose bins and the reported range of doses to the thyroid in HT patients. This was necessary as the original data only stated the incidence of HT for doses below 40 Gy.7 Excluding this data point did not significantly alter the fit. In 1 study, the dose to the thyroid was estimated as the prescription dose to the mantle fields,45 and in another study, the authors themselves estimated the thyroid dose from the mantle field prescription doses.32

Figure 3 shows the 4 dose-response fits, and Figure 4 shows the synthesized steepness, γ50, and position, D50, of the radiation dose-response curve. The fitted steepness of the dose-response curves varies from 0.56 to 2.3 with the synthesized value of γ50 = 1.4 (95% CI, 0.5-2.2), and the dose of 50% RIHT probability varied from 33 Gy to 65 Gy with a synthesized value of 45 Gy (95% CI, 28 Gy-62 Gy). Random effects modeling was used in the synthesis, as there was evidence of heterogeneity between the studies.

Figure 3.

Dose-response analysis of the 4 studies providing data amenable for this analysis and the overall estimate based on meta-analysis of the steepness and dose resulting in 50% complication of the 4 studies is shown (see Fig. 4).

Figure 4.

Meta-analysis of the steepness parameter of the dose-response curve, γ50, and the dose resulting in 50% risk of hypothyroidism, D50 is shown. The synthesized point estimates of D50 and γ50 parameterize the black dose-response curve of Figure 3.

DISCUSSION

HT incidence in the general population is well known to be higher among females than males. This increased risk is carried over in the case of RIHT, and it appears that RT works as a multiplicative risk factor, increasing the background risk of the general population.

Age does not appear to influence the risk of RIHT in the published studies. This is in contrast with other toxicities, for example, radiation-induced pneumonitis, where age has been shown to be a significant risk factor.48 Statistically, the width of the confidence interval of the OR is sufficiently narrow to rule out a clinically relevant age effect. However, as a note of caution, the study by Colevas et al20 did find an age effect in multivariate analysis (P = .004, the data used in this study), which was larger than in univariate analysis (P = .29 using 60 years as cutoff but P = .053 with age as a continuous variable). This suggests that older people do carry a higher risk of RIHT but that this risk is offset by more conservative prescription of RT. Our study cannot resolve such an effect because the majority of the data stem from univariate analyses. Also, it should be noted that all studies except Chin et al classify patients as young/old using a cut point of at least 20 years of age. Hence, it cannot be ruled out that age has an influence on the risk of hypothyroidism in pediatric patients.

Chemotherapy appears to have a negligible effect on the risk of HT. Again, the confidence interval of the OR essentially rules out a clinically significant effect. The actual drug, sequence, and dose intensity of the chemotherapy may be of importance, but because most of the included studies use multi-agent regimens, we suspect that an effect of a drug on the thyroid toxicity would have been seen as heterogeneity or a difference between tumor types. This is not evident in the data reviewed here.

The studies of Hancock and Koontz appear to be confounded—the former due to changing follow-up procedures along with changing use of chemotherapy, and the latter by a coupling between use of chemotherapy and the radiation dose prescribed.

Unsurprisingly, a partial thyroidectomy is associated with a highly increased risk of hypothyroidism. The data show, however, that even surgery not involving the thyroid gland is associated with an increase risk of RIHT, possibly due to increased vascular or nerve damage when the 2 modalities are combined. Heterogeneity among studies suggests that this effect depends on the exact surgical procedure.

For LAG, the pooled OR is significantly higher than unity, but there is heterogeneity and a tendency for the effect estimates to decrease with study weight. A formal test for publication bias yielded P = .097, which supports the suspicion of publication bias, especially considering that P = .1 is often recommended as the significance level in such tests.49 Questions over the reliability of the pooled estimate of the influence of LAG on RIHT mean that no firm conclusion can be drawn. However, the LAG analysis serves as an example of identification of bias in the meta-analysis and demonstrates the relevance of presenting data in a manner facilitating visual detection of suspected publication bias.

With regard to the relation between radiation dose and RIHT, the 4 studies included in the meta-analysis7, 23, 32, 45 as well as the 3 studies on dose response that reported insufficient data to be included in the meta-analysis1, 46, 47 all found evidence of a positive dose response. Thus, it seems safe to conclude that a higher radiation dose to the thyroid increases the risk of HT. Despite this, the fits presented in Figure 3 show major variability. Note that the 2 studies predicting a high radio-responsiveness, Kuten et al and Metzger et al, both included Hodgkin lymphoma (HL) patients only. It appears unlikely that HL patients are more susceptible to RIHT than head and neck cancer (H&N) patients, especially as age is not a risk factor. However, surgery, which is more likely to be performed in H&N patients, carries a high risk of HT. It is also conceivable that the symptoms of HT are more readily recognized in HL patients than in H&N patients. Furthermore, physicians may be more likely to focus on thyroid function in HL patients because the long-term side effects in this population are of major concern. Still, the grouping of HL and H&N patients in Figure 3 may also be a coincidence. We tried plotting the risk of HT versus radiation dose for all studies, allowing the extraction of just 1 data point and grouping the data by diagnosis. Although a tendency toward higher RIHT risk in HL patients than in H&N patients receiving the same dose could be seen from the plot, the overall variability in RIHT incidence was large, suggesting that the method and frequency of follow-up are more important than the tumor type (data not shown).

The length of follow-up is important because the incidence of RIHT has consistently been shown to increase with follow-up time within studies. However, patients in the study by Metzger et al had a median follow-up exceeding 11 years compared with those in Kuten at 1 to 6 years and Grande at 4 years, so length of follow-up alone cannot explain the differences seen.

Further complexity in the time evolution of RIHT may arise from a temporary increase in the TSH level followed by a reversion to normal levels, as reported by Nair et al34 in almost half the patients with elevated TSH. Such a transient TSH rise may lead to inflated RIHT incidence estimates in studies with an intense TSH measurement program and may explain some of the variation seen among studies. Hence, it may be relevant to define an HT end point requiring repeated measures of elevated TSH or, alternatively, grading HT according to the duration of the TSH rise.

In conclusion, the results of the dose-response studies are evidently not mutually consistent.

Our method of analysis parameterizes dose-response curves within each study and synthesizes the dose-response parameters across studies using inverse variance weighting. The result is a significant dose response, in concordance with the 7 studies referenced.1, 7, 23, 32, 45-47 However, the dose required to produce a 50% risk of RIHT is highly uncertain. Our study suggests the use of 45 Gy as an average number, but this is likely to be too high for the most rigorous methods of follow-up. Future studies correlating the length of TSH rise with dose are warranted to clarify the discrepancies shown in Figure 3.

Several studies have reported sparing of the thyroid by decreasing the volume of the thyroid irradiated. It appears that the thyroid shows a predominantly parallel structure, and the mean dose to the thyroid as used in this study or, alternatively, the volume receiving more than a certain threshold dose are the most promising metrics for predicting RIHT risk. Interestingly, the risk of radiation-induced thyroid cancer appears to decrease at sufficiently high doses,50 so a high dose to a small volume is most likely preferable to a low dose to a large volume in terms of both thyroid function and risk of second cancer.

A weakness of the current analysis is the use of odds ratios from univariate analysis for most of the studies. A confounding effect from differences in treatment between the 2 groups is therefore difficult to rule out. For example, as discussed above, older patients may be more conservatively managed than younger patients, and patients undergoing chemotherapy may receive a reduced radiation dose as in the study by Koontz et al.28

Another weakness of our study is that the incidence of HT increases with length of follow-up, and crude incidences are used in the calculations of the OR. Hence, if there are different follow-ups in 2 groups, the group with the shorter follow-up will appear artificially favorable. Although it is reasonable to assume that follow-up is independent of race, sex and age, it is more difficult to rule out an effect of follow-up versus chemotherapy, LAG, and surgery. This problem would have been avoided if the original studies had used survival statistics to report the incidence of late toxicity. Such data were not available in the vast majority of the included studies, but we encourage reporting of actuarial freedom from toxicity in a Kaplan-Meier analysis for late end points in future clinical studies.51 Reporting hazard ratios and CI in a multivariate Cox analysis circumvents both the above weaknesses in the current analysis.

Conclusions

Female gender and surgery, involving the thyroid gland or not involving the thyroid gland, are associated with a higher risk of RIHT. In addition, Caucasians are at higher risk of RIHT than African Americans. The meta-analysis of LAG showed an association with HT, but suspicion of publication bias weakens confidence in the result. There was clear evidence of a moderately steep dose response of radiotherapy, but the studies differed in the estimated dose, resulting in a 50% risk of RIHT, probably due to differences both in length of follow-up and diagnostic intensity.

CONFLICT OF INTEREST DISCLOSURES

I.R.V. is supported by The Lundbeck Foundation Center for Interventional Research in Radiation Oncology (CIRRO) and The Danish Council for Strategic Research. L.S. is supported by The Novo Nordisk Foundation, and S.M.B. acknowledges support from the National Cancer Institute grant 2P30 CA 014520-34.

Ancillary