Association Between Health Maintenance Practices and Skin Cancer Risk as a Possible Source of Detection Bias

Abstract

Importance

Detection bias may influence the results of epidemiologic studies of skin cancer risk. An individual’s degree of contact with the health care system, and, specifically, undergoing routine screening practices, may be a source of such bias. More intensive screening practices may be associated with increased diagnoses of basal cell carcinoma (BCC), squamous cell carcinoma (SCC), and melanoma.

Objective

To assess a possible association between health care screening practices and skin cancer risk.

Design, Setting, and Participants

The cohort of participants for this study was drawn from the Nurses’ Health Study (121 700 women) and Health Professionals Follow-up Study (51 529 men). Participants in the Nurses’ Health Study were followed up from June 1, 1990, to June 1, 2012, and participants in the Health Professionals Follow-up Study were followed up from January 1, 1990, to January 1, 2012. Statistical analysis was performed from April 4, 2017, to May 16, 2018.

Exposures

During cohort follow-up, Nurses’ Health Study and Health Professionals Follow-up Study participants were asked whether they had undergone various health care screening practices including physical examination by a physician, sigmoidoscopy or colonoscopy, eye examination, serum cholesterol test, mammography, breast examination and pelvic examination, and prostate-specific antigen test and rectal examination.

Main Outcomes and Measures

Incident BCC, SCC, and invasive melanoma. Cases of SCC and melanoma were confirmed with histopathologic findings. Hazard ratios (HRs) with 95% CIs were calculated for the association between screening practices and the various types of skin cancer.

Results

This study included 77 736 women from the Nurses’ Health Study (mean [SD] age at baseline, 56 [7] years) who were followed up for 1 388 523 person-years and 39 756 men from the Health Professionals Follow-up Study (mean [SD] age at baseline, 58 [10] years) who were followed up for 635 319 person-years. A total of 14 319 incident BCCs, 1517 SCCs, and 506 melanomas were identified in the Nurses’ Health Study cohort and 8741 incident BCCs, 1191 SCCs, and 469 melanomas were identified in the Health Professionals Follow-up Study cohort. Positive associations were seen between various screening practices and diagnoses of BCC and SCC, with similar directions of associations seen with melanoma for some screening practices. In the Nurses’ Health Study, the multivariable HR associated with undergoing a physical examination was 1.46 (95% CI, 1.30-1.64) for BCC, 2.32 (95% CI, 1.41-3.80) for SCC, and 1.66 (95% CI, 0.85-3.22) for melanoma. Similar results were seen in the Health Professionals Follow-up Study, with a multivariable HR associated with undergoing a physical examination of 1.43 (95% CI, 1.26-1.63) for BCC and 1.85 (95% CI, 1.17-2.92) for SCC, with an attenuated HR for melanoma of 1.04 (95% CI, 0.64-1.69).

Conclusions and Relevance

Undergoing health care screening practices increases the likelihood of being diagnosed with skin cancer. Researchers should be aware of this association and, where appropriate and possible, condition analyses of skin cancer risk on measures of health care use, including screening, to address confounding associated with detection bias.

This cohort study assesses a possible association between health care screening practices and skin cancer risk using data from the Nurses’ Health Study and Health Professionals Follow-up Study.

Key Points

Question

Does undergoing health care screening practices increase the likelihood of being diagnosed with skin cancer?

Findings

In this cohort study of 117 492 individuals from the Nurses’ Health Study and Health Professionals Follow-up Study, health care screening practices, including physical examinations performed by a physician, were associated with increased diagnosis of skin cancer, particularly basal cell carcinoma and squamous cell carcinoma.

Meaning

Researchers should be aware of this association and, where appropriate and possible, condition analyses of skin cancer risk on measures of health care use, including screening, to minimize the effect of confounding from detection bias.

Introduction

Detection bias is an important source of bias in epidemiologic studies. It arises when the likelihood of a disease being diagnosed (rather than caused) is different according to different levels of an exposure.¹ When researchers are interested in the causal effect of a disease or other correlate of intensity of medical care on an outcome, the possibility of detection bias must be addressed. This scenario is particularly true for scrutiny-dependent cancers (ie, cancers for which screening plays a role in detection), including skin cancer.² For example, in a hypothetical study of diabetes as a risk factor for basal cell carcinoma (BCC), people with diabetes may be more likely to be diagnosed with BCC because they see a physician regularly for diabetes-related physical examinations. Any increase in the risk of BCC associated with diabetes may result all or in part from patients’ interaction with the health care system, rather than from any etiologic link between the 2 diseases. To demonstrate the potential for detection bias in epidemiologic studies related to skin cancers, we examined the association between various health maintenance practices and diagnosis of squamous cell carcinoma (SCC), BCC, and melanoma in the Nurses’ Health Study (NHS) and the Health Professionals Follow-up Study (HPFS), large US cohorts of women (NHS) and men (HPFS).

Methods

Participants in the Nurses’ Health Study were followed up from June 1, 1990, to June 1, 2012, and participants in the Health Professionals Follow-up Study were followed up from January 1, 1990, to January 1, 2012, for incident SCC, BCC, and invasive melanoma. Follow-up in both cohorts occurs by biennial questionnaire, in which various health risk factors and diseases are assessed. Participants are periodically asked about health care screening practices, including physical examination performed by a physician, sigmoidoscopy or colonoscopy, eye examination, serum cholesterol test, mammography, breast examination and pelvic examination (NHS only), and prostate-specific antigen test and rectal examination (HPFS only). Participants are also asked if their physical examination was performed for screening purposes or to address specific symptoms. The studies did not ask about skin examinations. On the biennial questionnaires, participants are asked whether they have been diagnosed with SCC, BCC, or melanoma. For SCC and melanoma, all incident cases are confirmed by health record and review of pathologic findings. Record review is not performed for BCC, but previous validation studies have shown high rates of confirmation of BCC in these cohorts.^3,4 The NHS and HPFS are approved by the institutional review boards of Partners Health Care System and Harvard School of Public Health. The participants’ completion and return of the self-administered questionnaires implied informed consent.

For the analysis of each skin cancer, we excluded participants with a history of that skin cancer prior to 1990. Participants were followed up until their death; diagnosis of SCC, BCC, or melanoma; or the end of follow-up, whichever came first. Statistical analysis was performed from April 4, 2017, to May 16, 2018. We used Cox proportional hazards regression models to calculate age-adjusted and multivariable-adjusted hazard ratios (HRs) and 95% CIs for the association between various health maintenance practices and risk of SCC, BCC, and melanoma. For the analysis of each type of skin cancer, we additionally adjusted for previous diagnoses of the other 2 skin cancer types, as being diagnosed with 1 type of skin cancer increases the likelihood of being diagnosed with subsequent skin cancers. For physical examination performed by a physician, we additionally examined the exposure as a categorical variable (no examination, examination for screening purposes, and examination in response to symptoms). We studied the exposures as cumulative variables, where, once a participant has undergone that screening practice, he or she is considered exposed going forward. For physical examination performed by a physician, we used random-effects meta-analysis to calculate pooled HRs and 95% CIs between the 2 cohorts. We conducted a sensitivity analysis in which we additionally adjusted for various lifestyle covariates (body mass index; alcohol, caffeine, and citrus intake; and cigarette smoking). We conducted secondary analyses with melanoma in situ as the outcome. We performed secondary analyses for invasive melanoma stratified by body site of the primary tumor (head and neck vs other). We also examined our results stratified by calendar year (1990-1999 and 2000 and after). We used SAS, version 9.2 (SAS Institute Inc) for all statistical analyses. All statistical tests were 2-sided with significance defined as P < .05.

Results

In the NHS, we followed up with 77 736 women for 1 388 523 person-years and identified 14 319 incident BCCs, 1517 SCCs, and 506 melanomas. In the HPFS, we followed up with 39 756 men for 635 319 person-years and identified 8741 incident BCCs, 1191 SCCs, and 469 melanomas. The mean (SD) age at baseline was 56 (7) years in the NHS and 58 (10) years in the HPFS.

Overall, positive associations were seen in age-adjusted models between most health care screening practices and BCC and SCC, with more modest or no associations for melanoma (Table). Positive associations were generally attenuated in multivariable models; in the NHS, the multivariable HR associated with undergoing a physical examination was 1.46 (95% CI, 1.30-1.64) for BCC, 2.32 (95% CI, 1.41-3.80) for SCC, and 1.66 (95% CI, 0.85-3.22) for melanoma. Similar multivariable results were seen in HPFS for BCC (HR, 1.43; 95% CI, 1.26-1.63) and SCC (HR, 1.85; 95% CI, 1.17-2.92), with an attenuated HR for melanoma (1.04; 95% CI, 0.64-1.69). In the pooled analysis, undergoing physical examination performed by a physician was associated with BCC (HR, 1.45; 95% CI, 1.33-1.58) and SCC (HR, 2.05; 95% CI, 1.46-2.87) but not melanoma (HR, 1.24; 95% CI, 0.80-1.92). The results were consistent in sensitivity analyses adjusting for additional covariates. For melanoma in situ, the pooled risk associated with undergoing a physical examination was elevated, but imprecise (HR, 2.22; 95% CI, 0.90-5.45). The association of undergoing a physical examination with melanoma was not elevated for lesions on the head and neck, but was elevated for lesions off the head and neck for women (HR, 1.94; 95% CI, 0.91-4.13) more than men (HR, 1.23; 95% CI, 0.69-2.22). When stratifying by calendar year of follow-up (before 2000 vs 2000 or later), the results were generally stronger in 2000 or later (eTable in the Supplement).

Table. Cumulative History of Health Care Screening Practices and Risk of Melanoma, Squamous Cell Carcinoma, and Basal Cell Carcinoma.

Health Screening Practice	Basal Cell Carcinoma			Squamous Cell Carcinoma			Melanoma
	No. Exposed	HR (95% CI)		No. Exposed	HR (95% CI)		No. Exposed	HR (95% CI)
		Age Adjusted	MV Adjusteda		Age Adjusted	MV Adjusteda		Age Adjusted	MV Adjusteda
Nurses’ Health Studyb
Physical examination
Overall	14 032	1.51 (1.34-1.70)	1.46 (1.30-1.64)	1501	2.45 (1.49-4.02)	2.32 (1.41-3.80)	497	1.72 (0.88-3.34)	1.66 (0.85-3.22)
For screening	11 398	1.50 (1.33-1.68)	1.45 (1.29-1.63)	1225	2.43 (1.48-3.99)	2.31 (1.40-3.79)	411	1.73 (0.89-3.36)	1.68 (0.86-3.27)
For symptoms	2634	1.59 (1.40-1.79)	1.51 (1.34-1.71)	276	2.53 (1.53-4.20)	2.36 (1.42-3.91)	86	1.67 (0.83-3.33)	1.56 (0.78-3.13)
Colonoscopy or sigmoidoscopy	8841	1.31 (1.26-1.35)	1.28 (1.23-1.33)	1006	1.42 (1.27-1.59)	1.38 (1.23-1.54)	301	1.27 (1.05-1.53)	1.24 (1.03-1.50)
Eye examination	13 778	1.45 (1.31-1.60)	1.43 (1.29-1.58)	1479	1.52 (1.06-2.18)	1.47 (1.03-2.10)	481	0.99 (0.62-1.58)	0.98 (0.62-1.56)
Serum cholesterol test	12 691	1.48 (1.35-1.63)	1.45 (1.31-1.60)	1415	2.28 (1.60-3.24)	2.14 (1.50-3.06)	441	0.93 (0.62-1.39)	0.91 (0.60-1.38)
Mammography	13 614	1.66 (1.53-1.79)	1.61 (1.48-1.74)	1461	1.85 (1.40-2.43)	1.75 (1.33-2.31)	482	1.61 (1.05-2.45)	1.59 (1.04-2.43)
Breast examination	13 867	1.57 (1.43-1.74)	1.53 (1.38-1.69)	1491	2.54 (1.72-3.75)	2.41 (1.63-3.55)	483	1.01 (0.65-1.58)	0.99 (0.65-1.53)
Pelvic examination	13 963	1.44 (1.28-1.63)	1.39 (1.23-1.57)	1493	2.12 (1.36-3.30)	1.97 (1.27-3.07)	491	1.06 (0.60-1.89)	1.03 (0.58-1.83)
Health Professionals Follow-up Studyc
Physical examination
Overall	8495	1.47 (1.29-1.67)	1.43 (1.26-1.63)	1172	2.05 (1.30-3.23)	1.85 (1.17-2.92)	451	1.10 (0.68-1.77)	1.04 (0.64-1.69)
For screening	7407	1.47 (1.29-1.67)	1.44 (1.26-1.64)	1020	2.04 (1.29-3.22)	1.83 (1.16-2.90)	380	1.06 (0.66-1.72)	1.01 (0.62-1.64)
For symptoms	1088	1.45 (1.26-1.67)	1.42 (1.23-1.63)	152	2.11 (1.30-3.40)	1.91 (1.18-3.09)	71	1.30 (0.77-2.19)	1.22 (0.72-2.06)
Colonoscopy or sigmoidoscopy	6508	1.21 (1.15-1.27)	1.21 (1.15-1.27)	905	1.14 (1.00-1.31)	1.13 (0.98-1.29)	333	1.06 (0.86-1.31)	1.05 (0.85-1.29)
Eye examination	8293	1.28 (1.15-1.41)	1.23 (1.11-1.37)	1149	1.48 (1.07-2.03)	1.21 (0.87-1.68)	441	1.20 (0.79-1.83)	1.10 (0.72-1.69)
Serum cholesterol test	8456	1.44 (1.28-1.62)	1.40 (1.24-1.58)	1156	1.36 (0.97-1.90)	1.21 (0.86-1.70)	453	1.43 (0.86-2.36)	1.34 (0.81-2.23)
Prostate-specific antigen test	6484	1.45 (1.35-1.56)	1.42 (1.32-1.53)	982	1.79 (1.46-2.21)	1.64 (1.33-2.02)	329	1.21 (0.91-1.62)	1.15 (0.86-1.54)
Rectal examination	8159	1.33 (1.22-1.46)	1.30 (1.20-1.42)	1125	1.27 (0.98-1.63)	1.17 (0.91-1.51)	439	1.47 (1.01-2.15)	1.40 (0.96-2.05)
Pooled
Physical examination: overall	NA	NA	1.45 (1.33-1.58)	NA	NA	2.05 (1.46-2.87)	NA	NA	1.24 (0.80-1.92)

Abbreviations: HR, hazard ratio; MV, multivariable; NA, not applicable.

^aAdjusted for family history of melanoma, natural hair color, number of arm moles, sunburn susceptibility as a child/adolescent, number of lifetime blistering sunburns, and UV flux since baseline.

^bBasal cell carcinoma, 14 319 incident occurrences; squamous cell carcinoma, 1517; and melanoma, 506.

^cBasal cell carcinoma, 8741 incident occurrences; squamous cell carcinoma, 1191; and melanoma, 469.

Discussion

In this study focused on epidemiologic methods, we found positive associations between health care screening practices and skin cancer independent of skin cancer risk factors. Not all findings were statistically significant, but associations can introduce bias even in the absence of statistical significance. In general, more positive associations were seen for SCC and BCC. For melanoma, the association was more pronounced for lesions that were not on the head and neck. For obvious reasons, it is implausible that screening practices themselves cause an increased risk for skin cancer; our findings are not meant to draw any clinical conclusions. Rather, it is more likely that participants who actively engage with the health care system broadly, and undergo health screening practices in particular, will be diagnosed with skin cancers that would go undetected among those who are less engaged.

It is unclear why our findings were more pronounced for SCC, followed by BCC, then melanoma. Results for BCC may have been attenuated by our reliance on self-report: misclassification may have biased our results toward the null. Screening practices may have less of an association for melanoma, as these are often detected by patients themselves.⁵

Strengths and Limitations

Advantages of the cohort studies used for this analysis are their long duration of follow-up, validated data on skin cancer, and information on a wide variety of health care screening practices. However, given that our participants were nurses and health professionals, the findings in these populations may produce results that differ in the extent of bias from other, less medically sophisticated populations.

Conclusions

Our findings have important implications for skin cancer epidemiology research. Previous studies have found associations between various medical conditions and skin cancer risk without accounting for detection bias; these previous publications should be considered in that light because the associations might be confounded by health screening behavior.^6,7,8 Going forward, we encourage investigators studying skin cancer risk factors to account, whenever possible, for detection bias. Various strategies can be used, including conditioning on detection-related variables by restriction, stratification, or adjustment.¹ This is particularly imperative for studies on SCC and BCC and with exposures that can be easily linked with contact with the health care system.

Supplement.

eTable. Association of Having a Physical Examination by a Physician With Basal Cell Carcinoma, Squamous Cell Carcinoma, and Melanoma in Nurses’ Health Study and Health Professionals Follow-up Study Stratified by Calendar Year