CT Scans and Cancer Risks: A Systematic Review and Dose-response Meta-analysis

This systematic review and meta-analysis was successfully registered in the International Prospective Register of Systematic reviews (PROSPERO) on 28 April 2019, with the Registration number CRD42019133487. The initial enrollment was to study the relationship between CT screening and cancer risk in children and adolescents. We revised the registry for adult CT scans and cancer risks due to a published study by another group during our studying process [25]..

Observational studies, cohort studies, and case-control studies were included in this systematic review and meta-analysis, while studies assessed review articles, proceedings, case series and case reports were excluded. No restrictions to the language of publication were applied to select primary studies. We added extra studies by additional hand search of the reference lists of related articles.

Patients meeting the following requirements were included in this study: 1) Patients were older than 18 years for their first CT scan, 2) Patients did not have any malignant diseases (such as precancerous symptoms or early symptoms of the cancer) prior to their first CT scan, 3) Patients with follow-up of more than 1 year after their first scan. 4) Patients underwent at least one CT scan. Those with the following conditions were excluded: 1) Patients with incomplete demographic data or data errors, 2) Patients lost follow-up for a variety of factors.

Included patients received one or more electronically archived CT scans. The absorbed dose from a CT scan mainly depends on factors including age, sex, examination site, year of scan (2007 as a milestone for CT scans in terms of dose reduction in Annals of the ICRP [26]), and machine parameters.

The study included two outcome measures.

A thorough search was conducted in the following global databases, PubMed, Medline, EMBASE, Web of Science, Springer Link, Cochrane Library, from inception through Nov 15th 2020 with terms including “Adult”, “computerized tomography, X ray”, and “Cancer”. Search strategies were designed by an experienced librarian and revised by another librarian according to the PRESS (Peer Review of Electronic Search Strategies) checklist [27]. Three reviewers performed the selection process independently, and disagreements were discussed and resolved by a fourth reviewer.

The ROBs of each study were assessed independently by two authors according to Newcastle-Ottawa Scale [28]..

When missing data were encountered in included articles, we contacted the corresponding authors of relevant studies for these data by sending electronic mails twice with time interval of 1 month. When the study did not provide the number of cancer cases in the case/exposed group or control/unexposed group, the numbers were calculated by the formula defined as R_e/R_u for RR (Relative Risk), R_e (1- R_u)/[R_u (1- R_e)] for OR (the rate in exposed persons, denoted R_e; the rate in unexposed persons, denoted R_u), and (O_r/O_e)/(C_r/C_e) for HR (experimental group, denoted O; control group, denoted C; actual persons, denoted r; theoretical persons, denoted e) [29].

Publication bias were explored by a funnel plot (i.e., plots of study results against precision), Egger’s tests [30]..

In studies using LAR to estimate cancer risks, control group was defined according to the global incidence of cancer at 2.4 mSv of background radiation dose, i.e., lifetime baseline risk (LBR) [31]. We used the radiation risk models for sex- and organ-specific cancer incidence developed by the National Research Council’s BEIR VII committee [32]. LAR calculates the risk of cancer from an incubation period after radiation exposure (5 years for solid cancer and 2 years for leukemia) to survival to 100 years. LAR and LBR were calculated with the same method. For an age with a mantissa of 5: 45 years, divide by 2 the sum of the LAR values for 40 years and 50 years. For example, the LAR at the age of 45 years is (141 + 70)/2 = 105.5 at full 100 mSv (141 cases per 100,000 at 40-year-old and 70 cases per 100,000 at 50-year-old by BEIR VII). If less than 100 mSv, if it was 2.4 mSv, the LAR at the age of 40 years, 45 years, and 50 years are 41(2.4/100), 105.5(2.4/100), and 70(2.4/100), respectively.

Sensitivity analysis was performed by removing relevant studies to observe whether the homogeneity and the results change significantly. If the heterogeneity was too large to be analyzed, descriptive analyses were presented.

All analyses were performed using Stata V.16.0 software (Stata Corp LP, College Station, Texas, USA). Heterogeneity among primary studies were analyzed using standard Chi-squared tests (P value) and the Ι² statistic as recommended by the Cochrane Handbook for Systematic Reviews of Interventions [33]. We interpreted Ι² values according to Deeks [34]. Meta-regression was used to explore prior factors that may be important sources of heterogeneity. P < 0.05 was considered as a statistically significant.

The literature-retrieving strategy and pertaining results were shown in Fig. 1 and Table S1. A total of 44,657 relevant studies were preliminarily reviewed. In total, 125 studies satisfied the eligibility criteria for full-text screening and 25 were eventually included for this meta-analysis [2, 8‐11, 35‐54]. Of these 25 studies, 5 reported actual cancer cases [8, 10, 11, 35, 36], 19 estimated cancer cases by calculating LAR [2, 9, 37‐53], 1 reported expected cancer cases in control group [54], and all were published since 2007. Four studies were excluded due to incomplete data [12, 55‐57]. Additional data were requested by contacting the authors of 4 studies via repeated electronic mails with a time interval of 1 month (Table S2). However, no replies and requested data were received.

Studies were predominantly from the US (n = 11), followed by China (n = 4), Sweden (n = 3), Greece (n = 2), and Canada, Italy, Thailand, Croatia, and Denmark (n = 1 each); One study [42] included populations from China (Hong Kong) and the US. Based on the LAR base number of 100,000 persons, 25 studies were included with a sum of 111,649,943 participants, comprising 2,049,943 actual participants (111,449 cases VS 1938494 controls) from 6 studies; 109,600,000 estimated participants (54,912,051 participants received at least one CT radiation) from 19 studies. The average age was 51.3 years in CT scan exposed group (range, 20.0 to 94.0 years) and 40.5 years in control group (range, 25 to 84 years). The gender was reported in 19 studies of CT scan exposed group and 5 studies of unexposed group. The proportion of women was even higher (53.0%, 140,305 participants in CT group; 85.6%, 1,914,381 participants in unexposed group). The average follow-up time was 12.1 years in actual group (range, 5.9 to 22.5 years), 48.0 years (range, 5 to 80 years) in LAR group. The radiation dose per capita was 66.7 mSv (range:5.15 mSv to 122 mSv). A detailed list of study hallmarks was shown in Table 1 and Table 2.

The quality scores of 5 studies with actual participants and cancer cases ranged from 7 to 9 (Table S3A). Twenty of the studies were estimated cancer incidence and not case-control or cohort studies, the quality scores ranged from 2 to 6 (Table S3B).

The cancer risks for adults following CT scans were inordinately increased (LAR adults, OR, 10.00 [95% CI, 5.87 to 17.05], Ι² = 99.14; actual adults, OR, 1.17 [95%CI, 0.89 to 1.55], Ι² = 94.18; combined, OR, 5.89 [95%CI, 3.46 to 10.35], Ι² = 99.61%; Fig. 2). Notably, cancer incidence for actual studies was significantly different due to the inclusion (OR, 1.17 [95% CI, 0.89 to 1.55]) or exclusion (OR, 1.32 [95% CI, 1.11 to 1.56]) of Bruton’s study [35] with pregnant women participants (Table 3). Subgroup analysis indicated that cancer risks from LAR studies were significantly higher than actual studies (P = 0.00; Table 3), due to a longer life expectancy (to 100 years) used for calculation.

There was no evidence of publication bias (egger, P = 0.58). Cancer risks of exposed men and women were not significantly different (P = 0.71, men, OR, 4.89 [95% CI, 2.40 to 9.96]; Ι² = 99.43%; women, OR, 5.88 [95% CI, 2.96 to 11.68], Ι² = 99.52%; Table 3). There were no significantly different cancer risks (P = 0.93) for adults in terms of age (Table 3, less than 45 years, OR, 5.79 [95% CI, 2.08 to 16.14], Ι² = 99.64%; 45 to 65 years, OR, 5.04 [95% CI, 2.02 to 12.57], Ι² = 99.47%; older than 65 years, OR, 6.66 [95% CI, 2.07 to 21.38], Ι² = 98.00%).

The LAR estimation included the 2006 National Academy of Sciences BEIR VII. We extracted data in terms of scanning year (before and after 2007, 26] and countries (US or not) to explore the impact of scanning year and countries on cancer risks. Scanning year, countries, and continents of studies did not alter cancer risks (Ι² = 99.61%, P = 0.88; Ι² = 99.61%, P = 0.19; Ι² = 99.61%, P = 0.58, Table 3). There was no significant difference in study quality scores (Ι² = 99.61%, P = 0.25; Table 3).

Dose-response analysis showed that the cancer risks significantly increased linearly with radiation dose of CT scans (Coefficient = 0.03, P = 0.00) (Fig. S1). In details, the cancer risks were lowest for radiation dose of less than 15 mSv (OR, 2.51 [95% CI, 1.55 to 4.06], Ι² = 94.46%), highest for dose greater than 55 mSv (OR, 33.31 [95% CI, 21.33 to 52.02], Ι² = 97.09%), intermediate for dose of 15 to 55 mSv (OR, 5.48 [95% CI, 2.74 to 10.97]), with significant differences between triple groups (P < 0.001, dose grouping according to the overall radiation dose distribution included in the studies) (Fig. S2). Moreover, the relationship between cancer risk and radiation dose was more significant in the < 45 years group (P = 0.00), while there was no significant difference in 45–65 years groups (P = 0.08, Fig. S3).

For cancer types, a significantly higher risk of non-leukemic cancer was noted (P = 0.01; Table 3), in comparison with leukemic cancer. Cancer risks were significantly increased when more than 3 sites were scanned (P = 0.00; Table 3).

Sensitivity analyses did not alter the results, by excluding 1 study at a time from each meta-analysis (Table S4). Heterogeneity was present in all analyses (> 99% for all pooled estimates). Meta-regression (Table S5) identified data type, CT sites, and radiation dose contributing to heterogeneity for cancer risk, with no factors substantially reducing heterogeneity between studies.

The meta-analysis presents novel and complete evidence for the common and important issue concerned by clinicians, patients and the public, i.e., exposed CT scans and pertinent cancer risks for adults in their remaining lives. Based on 2,049,943 actual and 109,600,000 LAR/LBR participants from 9 countries, cancer risks for adults undergoing CT scans were successfully identified. Importantly, studies using LAR have been neglected in previous synthesized reports; partly due to the paucity of an appropriate control/unexposed population in quantify OR for these studies.

Indeed, selecting a control group (LBR) for LAR studies was challenging for such meta-analysis with global included studies. Participants unexposed to CT scans receive cosmic radiation (background radiation) in the earth, with various radiation dose range in different regions [31]. When background radiation was greater than 20 mSv per year, the area is designated as high natural background radiation (HNBR) region [58]. An ideal LBR should estimate cancer incidence based on various background radiation doses in different regions. Since HNBR areas were not covered in LAR studies for our meta-analysis, it is appropriate to use the global mean of 2.4 mSv per year as control group to estimate cancer risks. Our data showed that the incidence of cancer in the LBR group (42.7/100000 for men and 65.7/100000 for women) was lower than that in the CT exposed group (68.8/100000 for men and 91.9/100000 for women) and the updated annual global incidence of cancer [59]..

So far, no studies have quantified the cancer risk of low-dose radiation. One of the most important reasons is insufficient sample size. According to the evaluation of a total of 1000 mSv for a sample of 1000 people, the sample size needed to evaluate 100 mSv would be 100,000, 10 mSv would be 10 million [60]. In this meta-analysis, the sample size was sufficient with strengths, including 24,000,538 participants for CT radiation exposure < 15 mSv, 1,056,402,315 for 15 to 55 mSv, and 76,600,000 for dose > 55 mSv. Cancer risks increased with cumulative radiation dose from CT scans, consistent with previous evidence [61]. Meanwhile, the results showed that cancer risks increased slowly during radiation dose below 55 mSv, and rapidly for those above 55 mSv.

Ample evidence from Japanese atomic bomb survivors indicates that children are more susceptible than adults to the deleterious effects of ionizing radiation [62]. Children have longer life expectancy since exposure to ionizing radiation, providing more time for a cancer to manifest. This point supports our results with higher cancer risks for LAR than actual studies, since life expectancy used for LAR was 100 years [63]. Notably, there was no significant difference of age on cancer incidence when we divided adults into young, middle-aged and older (age groups according to the 2010 WHO Recommendations on physical activity for health) [64]. Similarly, age did not significantly alter cancer risks for children and adolescents exposed for CT scans in Huang’s study [25] (0 to 5 years (relative risk [RR], 1.35), 6 to 15 years (RR, 1.14), and > 15 years (RR, 1.24) [25]. Mavragani et al. found that DNA damage from low radiation doses was not severe enough to cause cell death, but could trigger a red flag [65, 66]. One of those red flags is the body’s immune system. Increasing evidence suggests that radiation exposure of low dose may not only have immunosuppressive effects, but likely to be associated with a radiation-induced enhancement in the immune system [67‐69]. However, immune systems of older adults are not as well functioning as those of younger people. This factor may underlie the results that the risk of cancer from CT radiation does not continue to decrease with age. This may also explain why cancer risks in the 45–65 years group are less sensitive to radiation exposure than those in the < 45 years group. Despite the age differences were insignificant, there was a significant increase in cancer risks for exposed adults compared with the background radiation group.

In 2007, the International Commission on Radiological Protection (ICPR) published a guide to the basic principles of appropriate radiation Protection based on radiation exposure. Since then, almost all international standards and national regulations dealing with radiation protection have been based on the recommendations of the Committee. Consequently, 2007 should be a milestone for CT scan in terms of dose reduction. Even in this study, no significant difference was found between radiation and cancer risk compared with studies published before or after 2007. What is clear is that researchers will adjust their protective strategies to a higher level in the plethora of exposure situations, as ICRP itself states.

Firstly, There are various confounders for cancer incidence [70], including genetics [71], hormone levels, [72, 73] tobacco, alcohol, overweight, physical activity [74, 75], socio-economic status, [76] and infectious agents [77, 78]. These confounders were not included in this meta-analysis, due to the unavailability of information from original studies. Secondly, the numbers of participants in this study are mostly derived from estimated group and tend to be theoretical cancer risks from CT scan. Nevertheless, evidence from these LAR studies should not be ignored for meta-analysis (have been ignored in previous systematic review). Thirdly, over 80% participants in this study were women, with the risk of cancer in women undergoing CT scans during pregnancy remains unclear. Fourthly, the actual and estimated participants were combined. The difference in age at the end of follow-up between the two groups may amplify the actual cancer incidence. Nevertheless, current follow-up time period from actual studies was relatively limited, with incomplete reflected cancer risks from CT scans. Fifth, it is apparent that CT scans were correlated with increased risk of solid cancers, i.e., cancer risk increased with radiation dose and sites scanned. However, the relative increased quantification of scans may be biased due to different weight factors of various organs/tissues of the body. Finally, heart disease was the most common reason for CT scans in included studies, with variation in terms of the severity of patients, the treatment plan, and the duration of follow-up, thus contributing to the inconsistency of included studies.