Association of Glycated Hemoglobin Levels With Risk of Pancreatic Cancer

Key Points Question What is the association of elevated glycated hemoglobin levels with the risk of pancreatic cancer? Findings In this cohort study with 851 402 participants, risk of pancreatic cancer varied in association with prior diabetes status as well as glycated hemoglobin level; however, this risk was not consistently observed across racial/ethnic groups. The number of patients who need to undergo further investigation to identify a single pancreatic cancer ranged from 206 to 600. Meaning In this study, the risk of pancreatic cancer associated with a newly identified elevation in blood glucose varied by race/ethnicity and did not reach a level sufficient to justify potential widespread screening.


Introduction
Pancreatic cancer is the third leading cause of cancer-related death in the United States, with a 5-year survival rate below 10%. 1 A contributor to poor survival for individuals who develop pancreatic cancer is the advanced stage of disease at the time of clinical presentation. The ability to detect pancreatic cancer at an earlier stage represents an important opportunity to potentially improve outcomes. However, because of the relatively low prevalence of pancreatic cancer (12.9 per 100 000 person-years [PYs]), 2 the United States Preventive Services Task Force does not recommend population-based screening. 3 As a result, targeted screening for patients from high-risk population subgroups has emerged as the most promising approach for early detection.
Hyperglycemia has been identified in patients with pancreatic cancer up to 36 months before cancer diagnosis. 4 Newly diagnosed (ie, incident) diabetes after the age of 50 years has received increasing attention as a potential marker of undiagnosed pancreatic cancer, 5 with 3-year cancer rates ranging from 0.25% to 1.0% [6][7][8] and the highest incidence reported in studies that have relied on glycemic criteria 7,8 rather than physician diagnosis. 6 However, it remains unclear to what extent glycemic parameters, particularly glycated hemoglobin (HbA 1c ) levels, are associated with overall risk of pancreatic cancer.
The objective of the present study was to evaluate the association of glycemic abnormality and risk of pancreatic cancer. We hypothesized that applying criteria other than those traditionally used in the definition of diabetes could lead to improved sensitivity for early detection of patients with pancreatic cancer. Therefore, we sought to characterize the performance of various thresholds of HbA 1c level, the most common test used to diagnosis diabetes, in a series of comparative cohort studies.

Methods
The present study was approved by the institutional review board of Kaiser Permanente Southern California (KPSC) with a waiver of informed consent because the study did not involve direct patient contact. This study is reported in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline. 9

Study Design and Setting
We performed a series of retrospective cohort studies to assess the association of varying glycemic criteria with the risk of pancreatic cancer. Analysis was performed on data collected from the research data warehouse of KPSC. Cohort entry was between January 1, 2010, and December 31, 2014, with follow-up through December 31, 2017. Kaiser Permanente Southern California is a community-based integrated health care system providing comprehensive care to 4.6 million enrollees. The full spectrum of health care services provided to health plan enrollees includes ambulatory care as well as acute hospital care, imaging services, and pharmacy services. The beginning of the study period was selected based on a shift in practice in favor of HbA 1c levels following revised guidelines from the American Diabetic Association that incorporated HbA 1c levels as part of the diagnostic criteria for diabetes. 10

Identification of Study Cohorts
We first identified patients aged 50 to 84 years who had at least 1 measure of HbA 1c between 2010 and 2014 in a KPSC medical facility (referred to as the base cohort). This age range was selected to identify a study cohort that would be representative of a potentially suitable population to undergo screening for early detection of pancreatic cancer. Blood samples collected in outpatient, emergency department, and inpatient settings were included. Point-of-care testing for HbA 1c levels was rarely performed during the study period and not included in the analysis. For each patient, the first laboratory test that met the entry criteria was referred to as the index HbA 1c test, and the date of the test was referred to as the index date. For the purpose of comparison, the following 12 contemporaneous cohorts were created. Patients with an HbA 1c level at or above prespecified thresholds (ie, 6.1%, 6.3%, 6.5% and 6.7% of total hemoglobin [to convert to proportion of total hemoglobin, multiply by 0.01]) during the study period, irrespective of previous values or diabetes status, were included in 4 separate elevated HbA 1c (EGH) cohorts, referred to as EGH 6.1% , EGH 6.3% , EGH 6.5% , and EGH 6.7% . These thresholds for HbA 1c level were selected a priori to provide a range of values across the spectrum of patients likely to be experiencing prediabetes or recent-onset diabetes. We created 4 diabetes-excluded cohorts (DECs), which included patients from each EGH cohort with no history of diabetes (extending back to 2000, when data were first available) before the index date based on a previously validated algorithm for identification of patients with diabetes.
This algorithm defined diabetes based on diagnosis codes, use of diabetes medication, or elevated glycemic laboratory values. We used the following logic to identify and exclude patients with history of diabetes: any hospital discharge code for diabetes (International Classification of Diseases, Ninth Revision [ICD-9] code 250.XX), any KPSC internal code for diabetes (ie, 200, 1201, 1202, 1203, 1204, 1839, 3141, 3186, 3639, 4124, or 5782), any prior HbA 1c level greater than 7.0%, or any dispensing record for insulin or an oral hypoglycemic medication (not including metformin). These cohorts were designated DEC 6.1% , DEC 6.3% , DEC 6.5% , and DEC 6.7% . Finally, we created confirmed index hyperglycemia cohorts (IHCs). To evaluate a potential increase in positive predictive value by adopting more stringent criteria defining patients with new-onset hyperglycemia, we established an additional series of patient cohorts including only patients with a confirmed finding of newly elevated HbA 1c . Confirmation of newly elevated HbA 1c level was performed by requiring a test below the respective threshold in the 18 months before the index date. The criteria defining IHCs were selected to reflect those of an ongoing prospective cohort study. 11 These cohorts were designated IHC 6.1% , IHC 6.3% , IHC 6.5% , and IHC 6.7% . For all study cohorts, patients with history of pancreatic cancer before the index date either based on diagnosis code (ICD-9 code 157.x or ICD-10 code C25.x) or KPSC Cancer Registry 12 and patients not enrolled in the health plan on the index date were excluded.

Outcome Definition
The primary outcome, pancreatic cancer, was ascertained through the KPSC Cancer Registry 12 and linkage of enrollees with decedents from the California death files. 13 The cases identified through the Cancer Registry were pancreatic ductal adenocarcinoma (PDAC) based on specific histology codes.
However, such a restriction was not applied to the cases from the California death files because the information on histology was not available.

Statistical Analysis
For each of the 12 cohorts, the following analyses were performed. We first estimated the 3-year risk and rate of pancreatic cancer per 1000 PYs and their 95% CIs. Then we calculated the number of patients needed to screen to detect 1 PDAC case. Among patients diagnosed with or who died of pancreatic cancer, we examined time to cancer with 2 different approaches, as follows: median (interquartile range [IQR]) follow-up time and cumulative percentage diagnosed or died in the first, second, and third years. Finally, we reported the distribution of cancer stage at the time of diagnosis.
Analyses were conducted in SAS statistical software version 9.4 (SAS Institute). All analyses were descriptive in nature. We did not perform hypothesis testing and thus did not prespecify a level of statistical significance.

Eligible Study Participants
Among the base cohort of 851 402 patients, 447 502 (52.5%) were women, 255 441 (30.0%) were Hispanic participants, 383 685 (45.1%) were non-Hispanic white participants, 100 477 (11.8%) were Asian participants, and 88 969 (10.4%) were non-Hispanic black participants, with a median (IQR) age of 62 (56-69) years and a median (IQR) HbA 1c level of 6.0% (5.7%-6.6%) ( Table 1). The number of unique eligible study participants included in the EGH cohort, DEC, and IHC with varying HbA 1c thresholds can be found in the Figure. After excluding prior diabetes as well as confirmation of new-onset hyperglycemia based on an HbA 1c level of 6.5%, a total of 20 012 patients remained in the IHC 6.5% cohort, with 74 of 1041 PDAC cases (7.1%) from the base cohort included. The cohort size ranged from a high of 495 310 patients (58.2% of the base cohort) for EGH 6.1% to a low of 12 540 (1.5% of the base cohort) for IHC 6.7%.

Patient Characteristics
Patients in the EGH cohort and DEC had similar median ages (61-62 years) compared with the base cohort (62 years of age) (Table 1). However, patients in the IHC tended to be 1 to 2 years older (median age, 63-64 years) compared with patients in the base cohort, EGH cohort, and DEC (Table 1).
In addition, the IHC appeared to have similar or higher frequency of women (52.1%-55.1%) compared with patients in the base cohort (52.5%). There was variation in racial/ethnic composition across the

Risk and Rate of PDAC
In the base cohort, the 3-year risk and the incidence rate of pancreatic cancer were 0.12% (95% CI, 0.12%-0.13%) and 0.45 (95% CI, 0.43-0.49) per 1000 PYs, respectively. The estimated 3-year risk and incidence rates per 1000 PYs for each cohort are reported in

Number and Proportion of PDAC Cases
The total number PDAC cases in the base cohort was 1041.  Table 4 presents the estimated number of cases that would need to be evaluated in a 3-year period to detect a single case of pancreatic cancer based on the cancer risk observed within each of the study cohorts (assuming 100% ability to identify an existing cancer). The estimate ranged from a low of 206 (95% CI, 160-264) in patients in IHC 6.7% to a high of 600 (95% CI, 540-666) in patients in the EGH 6.1% cohort. The number needed to undergo evaluation to identify a single case of PDAC in the base cohort was 818 (95% CI, 770-869).

Discussion
In this large retrospective comparative cohort study, the risk of pancreatic cancer in persons aged 50 to 84 years varied according to both the degree of elevation in HbA 1c level as well as timing with respect to onset of diabetes. Adoption of specific thresholds for elevation in HbA 1c level while applying the broadest set of cohort entry criteria without exclusion based on prior diabetes status provided the greatest sensitivity (range, 62%-80%) for detection of pancreatic cancer while still affording a 1.4-to 1.8-fold increase in cancer incidence from the base cohort. Using the same thresholds for HbA 1c level but applying a much more stringent definition for incident hyperglycemia yielded a 1.7-to 4.0-fold increase in cancer incidence at the expense of sensitivity (range, 6%-9%).
A major contributing factor to poor survival in pancreatic cancer is the late stage at diagnosis. 14 Based on the relatively low incidence, it is unlikely that widespread, population-based screening would prove beneficial. However, a targeted approach to screening patients at increased risk offers the potential for early detection, thereby improving survival. While efforts to apply screening in patients at increased risk based on family history or genetic predisposition appear promising, 15,16 these inherited forms account for a very small percentage (ie, 3% to 5%) of pancreatic cancer cases.
Therefore, attention has turned to evaluating a broader segment of the population. In particular, there is a growing body of evidence suggesting that patients aged 50 years or older with incident diabetes constitute an additional high-risk population for pancreatic cancer. [17][18][19] While previous attempts have focused on identifying a high-risk subgroup based on the identification of new-onset diabetes, 6,8 the present study expands on previous research by providing estimates of pancreatic cancer incidence across a range of newly established hyperglycemia, including levels that would qualify as prediabetes. The present study also highlighted potential differences in the association of new-onset hyperglycemia and the risk of pancreatic cancer based on race/ethnicity that have not been reflected in recent attempts to develop risk-prediction models to further identify high-risk subgroups. Accounting for these differences in future approaches to risk stratification is an important step to avoid potentially exacerbating existing disparities in pancreatic cancer. [20][21][22] The observed rates of pancreatic cancer among patients with hyperglycemia were lower than prior estimates based on patients with new-onset diabetes. Population-based estimates of pancreatic cancer incidence among individuals with new-onset diabetes have focused on the first 3 years after diagnosis. These estimates have varied from 0.4% to 1.0%. 6,8 An explanation for the discrepancy has been that studies incorporating diabetes diagnosed based on glycemic parameters tended to report increased rates of pancreatic cancer compared with reliance on diagnosis codes, potentially because of a delay in clinical diagnosis despite abnormal glucose or HbA 1c values. 23 Although we used glycemic criteria in the present study, estimates for 3-year risk of pancreatic cancer were closer to 0.4% among patients with confirmed incident diabetes using a threshold of 6.5% for HbA 1c level. There are several potential explanations for the lower estimate observed in the present study. First, we focused exclusively on HbA 1c levels as the glycemic parameter of interest, rather than on fasting glucose, which was the predominant measure used in previous studies reporting higher rates of pancreatic cancer. 8,17 We focused on HbA 1c level as a measure of hyperglycemia because of the increased use of this parameter in the KPSC health system following the 2010 guideline from the American Diabetes Association, which included elevation in HbA 1c level as part of the diagnostic criteria for diabetes. 10 Second, the current study population was substantially more racially/ethnically diverse compared with those in prior studies. 7,8,24

Limitations and Strengths
There were several important limitations to the present study. First, as previously noted, we limited the analysis to evaluation of HbA 1c level. It is conceivable that cancer rates would vary if other measures of hyperglycemia were included. Second, we did not examine the role of antihyperglycemic medications, which could have played a role in determining glycemic status in the study cohorts that included patients with a history of diabetes. This was beyond the scope of the study, given that exposure to antidiabetic medications was an exclusionary factor for most of the cohorts included in the analysis. Additionally, selection bias may have occurred based on clinicianrelated decisions to measure HbA 1c level, given that patients with this measure may have had poorer health at baseline or been more likely to have established diabetes (as evidenced by the relatively elevated median HbA 1c level in the base cohort). As a result, we pursued a variety of methods to exclude patients with evidence of prior diabetes diagnosis from the analytic study cohorts. In addition, because cases from the California death file did not have histologic diagnoses available, it is possible that a small proportion of non-ductal adenocarcinoma subtypes of pancreatic cancer may have been included in the analysis. Furthermore, while we have provided an estimate of the number of patients needed to undergo investigation to detect a case of pancreatic cancer under a variety of scenarios, it was not possible to provide a more precise estimate of the number needed to screen given lack of data on the potential effectiveness of currently available early detection strategies based on either cross-sectional imaging or endosonography to improve survival in pancreatic cancer. [25][26][27] Despite the study limitations, the present study has multiple strengths. These include the relatively large sample size and racially/ethnically diverse study population. In addition, the setting of an integrated health care system enabled accurate assessment of prior diabetes status. Finally, the use of a prospectively maintained cancer registry as well as a state-wide death index further ensured accurate identification of patients who developed pancreatic cancer.

Conclusions
In this study, elevated HbA 1c level among individuals aged 50 to 85 years was associated with increased risk of pancreatic cancer. However, the risk varied by race/ethnicity, with the highest risk noted among non-Hispanic white patients with confirmed evidence of incident hyperglycemia in the absence of prior diabetes or elevated HbA 1c value. An increased risk of pancreatic cancer based on more stringent definitions of new-onset hyperglycemia was not observed among other racial/ethnic groups. Ultimately, the number of patients needing to undergo testing based solely on elevation in HbA 1c level exceeded reasonable limits based on available testing strategies and resources.
Alternative approaches to risk stratification are still needed to improve early detection of pancreatic cancer.