Surveillance Bias in Cancer Risk After Unrelated Medical Conditions: Example Urolithiasis

We analysed cancer risks in patients with urinary tract stones but some features of the generated results alarmed us about possible surveillance bias, which we describe in this report. We used nationwide Swedish hospital records to identify patients with urinary tract stones (N = 211,718) and cancer registration data for cancer patients for years 1987 to 2012. Standardized incidence ratios (SIRs) for cancer were calculated after the last medical contact for urinary tract stones. All cancers were increased after kidney (SIR 1.54, 95%CI: 1.50–1.58), ureter (1.44, 1.42–1.47), mixed (1.51, 1.44–1.58) and bladder stones (1.63, 1.57–1.70). The risk of kidney cancer was increased most of all cancers after kidney, ureter and mixed stones while bladder cancer was increased most after bladder stones. All SIRs decreased steeply in the course of follow-up time. Tumour sizes were smaller in kidney cancer and in situ colon cancers were more common in patients diagnosed after urinary tract stones compared to all patients. The results suggest that surveillance bias influenced the result which somewhat surprisingly appeared to extend past 10 years of follow-up and include cancers at distant anatomical sites. Surveillance bias may be difficult to avoid in the present type of observational studies in clinical settings.

The early aim of the present study was to assess the risk for urological cancers in patients who had previously been diagnosed with UL. UL is a common disease affecting up to 15% of population and many patients have a recurrent disease 8,9 . UL is thought to be associated with the risk of kidney and bladder cancers, and a study from Taiwan's National Health Insurance Research Database reported that UL is associated with a high risk of many systemic cancers in addition to urinary tract cancers [10][11][12] . Thus we considered analysing all cancer types after UL diagnosis. In the present study we use nation-wide Swedish hospital records of inpatients and outpatients with UL diagnoses and link the individual data to cancer data. We analyse the data in terms of follow-up time, proportion of in situ tumours and tumour size. In the course of the analyses the likelihood of detection bias became apparent, and the aims were shifted towards featuring some typical signs of such a bias.

Results
The total number of patients diagnosed with UL during years 1987 to 2012 was 211,718, distributed by the most common type, ureter stones (91,397, 43.2%), followed by kidney stones (77,972, 36.8%), mixed stones (23,890, 11.3%) and bladder stones (18,459,8.7%). We analysed the risk of 8 specific cancers, all other cancers together and all cancers combined after the last identified medical contact for the four types of UL (Table 1). All cancers were rather uniformly increased after kidney (SIR 1.54), ureter (1.44), mixed (1.51) and bladder stones (1.63). The risk of kidney cancer was increased most after kidney (3.75), ureter (2.78) and mixed stones (4.01) while bladder cancer was increased most after bladder stones (3.34). After bladder stones, SIRs were highest for prostate (1.74) and colon (1.60) cancers. Remarkably, all the listed cancers were increased in patients with kidney and ureter stones. Cancers at anatomically distant sites, in the lung and the breast, were increased after all types of UL, except for lung cancer after bladder stones.
In Table 2 cancer risk after any UL diagnosis was assessed by follow-up time since the last UL diagnosis. For all cancer the SIR was 4.70 during the first year (4239 cases) but then stabilised to 1.31 (12,159 cases) and 1.24 (5208 cases) during 1-9 and 10 + years of follow-up, respectively. The initial SIRs were excessive, 22.40 and 14.02 for kidney and bladder cancers but they were high also for prostate (4.89) and pancreas (4.80) cancers. All but lung cancers were increased even at the latest follow-up time: risk for kidney cancer remained the highest (1.56) while non-urological cancers showed SIRs between 1 1.09-1.16); 20.8% of the patients had at least one visit, and their cancer risk was not increased (0.98, 0.93-1.14).

Discussion
Based on the results in Table 1 a noncritical interpretation would point out that UL is associated with cancer at multiple sites, even though the affected organs appeared to convey the highest risk. Considering the high prevalence of UL in the population, a logical conclusion should have been that UL is associated with a high population burden for cancer. The follow-up results in Table 2 exposed the concerns about surveillance bias, which may to be due to incidental clinical findings by computed tomography in diagnosis of kidney, ureter and mixed stones. Bladder stones are diagnosed by cystoscopy and palpation of the prostate, which may explain the high risks of prostate and colon cancers. All UL patients may be imagined by thorax radiography which may reveal tumours in the covered parts of the body.
It was somewhat surprising that the bias appeared to extend past 10 years of follow-up. The reason may be that in many patients UL is recurrent or at least may increase concerns leading to frequent medical contacts 8,9 . However, using the Outpatient data to follow UL patients for multiple UL related visits, only 20.9% had one or more visits, suggesting that many UL patients with multiple episodes were seen in the primary care for which we have no data. Also the data showed that the simple number of Outpatient visits for UL causes did not appear to explain cancer risks because even patients with no additional visits showed an excess risk of 1.13.
The more frequent in situ diagnoses and smaller tumour sizes In UL patients compared to all cancer patients supported the existent of surveillance bias although the differences were modest. The mechanism of UL related cancer, in humans as well as in experimental animals, is believed to be mechanical wear and inflammation [10][11][12] . Yet it would not be self-evident that UL would cause systemic cancers because usually inflammation, for example, by autoimmune diseases is associated with cancers at sites affected by inflammation [13][14][15][16] .
Although the above data suggest involvement of surveillance bias, we need to admit that we have not eliminated alternative mechanisms which predispose to UL and which might jointly predispose to cancer. Stone formation is due to a combination of genetic and environmental factors. Risk factors include high urine calcium levels, obesity, certain foods, some medications, calcium supplements, hyperparathyroidism, gout, diabetes, hypertension and not drinking enough fluids 9 . Among these obesity and diabetes are known risk factors of cancer, and kidney cancer, showing the highest risk in the present study, is one of the highest obesity-related cancers [17][18][19] .
Whether it is possible to eliminate surveillance bias by a study design may not be simple. A frequently applied method is the omission of early follow-up but this may lead to removal of true cases if these are properly diagnosed, such as histologically verified cancers. Comparison of cancer risk between early and late onset UL and in persons suffering from multiple episodes should be informative but not bias-free. We conclude that quantification of risks may be extremely difficult in situations when surveillance bias is likely, and scientists and readers of scientific texts should be aware of such problems in observational studies.

Methods
UL patients were identified using the nationwide Swedish Hospital Discharge Register (1986-2012) and the Outpatient Register (2001-2012). The first UL diagnosis in either register was included and a patient was only entered once. Information from the registers was linked at the individual level via the national 10-digit civic registration number to the Swedish national Cancer Registry. In the linked dataset, civic registration numbers were replaced with serial numbers to ensure the anonymity of all individuals. Revisions 9 (1987Revisions 9 ( -1996 and 10 (1997-) of the International Classification of Diseases (ICD) was used to identify UL diagnostic codes. Only 54,500 patients were diagnosed during the ICD-9 period, compared to 166,600 in the ICD-10 period. Standardized incidence ratios (SIRs) were calculated as the ratio of observed to expected number of cases. The follow-up was started from the last UL diagnosis and ended at diagnosis of the first primary cancer, death (Causes of Death Register) or end of follow-up, December 31, 2012, whichever came first. The expected numbers were calculated for all individuals without a history of UL (i.e., essentially for the whole Swedish population covered by the Database), and the rates were standardized by 5-year-age, gender, period (5 years group), socioeconomic status and residential area. The 95% confidence interval (95%CI) of the SIR was calculated assuming a Poisson distribution. Data on the TNM classification were available from year 2002 onwards.
In order to assess the frequencies of medical contacts due to UL and their influence on SIRs we identified patients who were diagnosed with UL using the Hospital Discharge Register during 1987 to 1999 and followed their UL related visits from the Outpatient Register during 2001-2012.
Ethical statement. The study was approved by the Ethical Committee of Lund University and the study was conducted in accordance with the approved guidelines not requesting informed consent. The study is national register-based study on anonymous personal data.