Abstract
Early-detection tests for pancreatic ductal adenocarcinoma (PDAC) are needed. Since a hypothetical screening test would be applied during antecedent clinical encounters, we sought to define the variability in health-care utilization leading up to PDAC diagnosis. This was a retrospective cohort study that included patients diagnosed with PDAC in the Indianapolis, Indiana, area between 1999 and 2013 with at least 1 health-care encounter during the antecedent 36-month period (n = 1,023). Patients were classified by unique patterns of health-care utilization using a group-based trajectory model. The prevalences of PDAC signals, such as diabetes mellitus (DM) and chronic pancreatitis, were compared. Four distinct trajectories were identified, the most common (42.0%) being having few clinical encounters more than 6 months prior to PDAC diagnosis (late acceleration). In all cases, a minority of persons had DM (30.6%, with 9.5% <1.5 years before PDAC) or any pancreatic disorder (39.9%); these were least common in the late-acceleration group (DM, 14.7%; any pancreatic disorder, 32.1% (P < 0.001)). The most common pattern of antecedent care was having few clinical encounters until shortly before PDAC diagnosis. Since the majority of patients diagnosed with PDAC do not have an antecedent PDAC signal, early-detection strategies limited to these groups may not apply to the majority of cases.
Keywords: carcinoma, ductal; diabetes mellitus; pancreatic neoplasms; pancreatitis, chronic
Pancreatic ductal adenocarcinoma (PDAC) is the fourth leading cause of cancer death, and its dismal prognosis has changed little in recent decades (1). There is substantial interest in identifying at-risk populations who may benefit from early-detection strategies. However, its incidence is sporadic in the majority of patients (2). Since widespread screening for PDAC is currently impractical, several risk factors have been considered to enrich a population who might benefit from early PDAC screening. These include family history of relevant cancers, personal history of alcohol drinking and smoking, pancreatic cystic neoplasms, chronic pancreatitis, and diabetes mellitus (DM). The incidence of PDAC within groups having 1 or more of these risk factors is too low to warrant population-based screening. Furthermore, screening cannot be instituted for patients who do not utilize the health-care system until the onset of overt PDAC-specific symptoms and signs, such as unrelenting abdominal pain and obstructive jaundice. Thus, patients must have 1 or more health-care encounters well in advance of their PDAC diagnosis in order for known and future PDAC risk factors to be identified and prompt its earlier detection. Therefore, given the substantial interest in early detection, it is important to define the epidemiologic variability in clinical presentations with PDAC.
We hypothesized that persons diagnosed with PDAC could be distinguished on the basis of unique trajectory patterns of their health-care utilization in the years leading up to a diagnosis of PDAC. Patients with greater utilization of the health-care system years prior to their PDAC diagnosis would be a “captive audience” for earlier detection; on the other hand, patients who only entered the health-care system closer to their cancer diagnosis would be more challenging to diagnose at an earlier stage. Therefore, our aims in this study were to define unique patterns of health-care utilization in the 36-month period preceding a diagnosis of PDAC and to identify patient characteristics associated with these distinct patterns.
METHODS
Study design and population
This was a retrospective cohort study of patients diagnosed with PDAC between January 1999 and December 2013 in the Indianapolis, Indiana, metropolitan area. Patients were identified using clinical and insurance claims data derived from the Indiana Network for Patient Care (INPC), a nationally recognized regional health information exchange that includes data from 25,000 physicians, 106 hospitals, 110 clinics and surgery centers, and other health-care organizations, including insurance carriers, across Indiana (3). The network includes information from encounters covering more than 90% of care provided at hospitals in the Indianapolis area. The study was approved by the Indiana University Office of Research Administration before commencement of data collection.
Enrollment criteria and definitions
All diagnoses were based on clinical encounters classified according to the International Classification of Diseases, Ninth Revision (ICD-9). Patients aged ≥18 years who were diagnosed with PDAC, defined by at least 2 clinical encounters associated with a relevant ICD-9 code (157.x), were included. Patients who were not included in the Indiana State Cancer Registry (http://www.in.gov/isdh/24968.htm) were excluded. Patients who resided in counties outside of the Indianapolis metropolitan area (Marion County and its 8 surrounding counties) were excluded, to minimize detection bias in our assessment of antecedent health-care utilization. Patients with a concurrent cancer diagnosis (ICD-9 codes 140–156.0, 158.0–208, and 230–239) other than nonmelanoma skin cancer (ICD-9 code 173.x) were also excluded. In order to evaluate the prevalence of potential risk factors (and signals or harbingers of PDAC) prior to the diagnosis of PDAC, we excluded patients with no antecedent encounters in the INPC. When available, PDAC stage at diagnosis was defined using INPC data linked to the Indiana State Cancer Registry.
For the purposes of this study, PDAC signals included DM (ICD-9 codes 249.x and 250.x), acute pancreatitis (ICD-9 code 577.0), chronic pancreatitis (ICD-9 code 577.1), and other pancreatic disorders (defined as ICD-9 codes 577.x, other than codes 577.0 and 577.1).
Health-care utilization
Health-care utilization was defined by health-care encounters that occurred during the 36-month period prior to the diagnosis of PDAC. Health-care encounters were further subdivided on the basis of type: inpatient, outpatient, and emergency room. Each encounter was then defined as related/potentially related to PDAC based on its associated ICD-9 code(s). A full list of related/potentially related ICD-9 codes is provided in Web Table 1 (available at https://academic.oup.com/aje); these included symptoms such as abdominal pain (ICD-9 codes 789.00–789.09, 789.60–789.69, and 789.7), weight loss (ICD-9 codes 783.21, 783.22, 783.41, and 783.7), DM (ICD-9 codes 249.x and 250.x), diseases of the pancreas (ICD-9 code 577.x), and bile duct obstruction (ICD-9 code 576.2), among others. Any clinical encounter that did not contain 1 or more of these related/potentially related codes was classified as unrelated. Since each encounter could contain 1 or more ICD-9 codes, these encounter-level data are presented on a per-indication basis.
Statistical analysis
In order to examine the longitudinal trend of health-care utilization, we evaluated whether 1 or more encounters occurred in each of the 12 quarterly intervals during the 36-month period prior to PDAC diagnosis. We considered related/potentially related and unrelated encounters in this analysis. Based on these data, subgroups of patients with distinct trajectories of health-care utilization were jointly identified using inpatient, outpatient, and emergency-room encounters via the group-based trajectory model (4–6). The group-based trajectory model is a semiparametric model with the assumption that the population consists of multiple trajectory groups, defined on the basis of 1 or more longitudinal measures.
Since there were 3 outcomes of interest, including inpatient, outpatient, and emergency-room service utilization, a multitrajectory model was used in which each trajectory group was defined by 3 longitudinal trajectories. The group-based trajectory model was fitted using SAS PROC TRAJ (SAS Institute, Inc., Cary, North Carolina) with the logistic model due to the binary nature of the data (7). To determine the number of trajectory groups, we evaluated models with 1–6 trajectory groups with cubic trajectories and compared the fits of the models using the Bayesian Information Criterion. The number of trajectory groups was chosen on the basis of both the Bayesian Information Criterion and adequate group size, with at least 10% of the sample chosen for each trajectory group (8). Once the best-fitting model had been identified, we then compared the cubic trajectory model with quadratic and linear trajectory models to determine the final model. The posterior predictive probabilities that a patient belonged to each of the trajectory groups were computed based on the final model. Patients were then assigned to the most likely trajectory group that best fitted their encounter pattern.
Patient characteristics at the time of PDAC diagnosis were summarized using mean values and standard deviations for normally distributed continuous variables, median values and interquartile ranges for non–normally distributed continuous variables, and frequencies and proportions for categorical variables. Comparison of patient characteristics across the trajectory groups was performed using analysis of variance for normally distributed continuous variables, the Kruskal-Wallis test for non–normally distributed continuous variables, and Pearson's χ2 test for categorical variables. The associations between patient characteristics and trajectory groups were evaluated using a multinomial logistic regression model with a generalized logit link. All statistical analyses were performed using SAS, version 9.4.
RESULTS
Variability in health-care utilization leading up to the diagnosis of PDAC
Between January 1999 and December 2013, there were 3,374 patients diagnosed with PDAC in the INPC registry who had data available in the Indiana State Cancer Registry. Of these persons, 2,239 (66.4%) resided outside of the Indianapolis metropolitan area and 3 (0.01%) were under 18 years of age, leaving 1,132 incident cases of PDAC. Of these patients, 109 (9.6%) had had no health-care encounters in the 36-month period prior to the diagnosis of PDAC, leaving for study inclusion 1,023 patients with 1 (n = 98; 9.6%), 2 (n = 82; 8.0%), or ≥3 (n = 843; 82.4%) health-care encounters.
We then examined the longitudinal trends in health-care utilization during the 36-month period leading up to the diagnosis of PDAC. Using encounter-level data that included all encounters irrespective of their relatedness to the ultimate cancer diagnosis, we identified 4 distinct trajectories for patients’ pre-PDAC clinical course (Figure 1). Although the 5-group model provided a slightly better fit to the data, we chose the 4-group model because the improvement in Bayesian Information Criterion from 4 groups to 5 groups was substantially smaller than that from 3 groups to 4 groups. Persons in trajectory group 1 (late acceleration, 42.0%) had minimal numbers of clinical encounters until the 2 quarters immediately prior to PDAC diagnosis; group 2 individuals (early acceleration, 9.7%) had a pattern of increasing health-care utilization throughout the 36-month period; group 3 individuals (high outpatient utilization, 27.0%) had consistently high outpatient utilization but minimal inpatient/emergency room utilization; and group 4 individuals (high overall utilization, 21.3%) had consistently high utilization of all encounter types during the 36-month period. The proportion of women, the proportion of nonwhites, and baseline comorbidity increased significantly from group 1 to group 4 (Table 1; P < 0.001 for each). With the exception of the late-acceleration group (group 1; 34.4%), more than 97% of individuals represented in groups 2–4 had at least 1 health-care encounter 1 or more years prior to their PDAC diagnosis, during which PDAC signals or the cancer diagnosis could have been identified.
Figure 1.
Patterns of health-care utilization during the 36 months prior to diagnosis of pancreatic ductal adenocarcinoma (PDAC), Indianapolis, Indiana, 1999–2013. Using group-based latent trajectory analysis, the graph illustrates 4 unique patterns of health-care utilization in the 36-month period leading up to a PDAC diagnosis: late acceleration (A), early acceleration (B), high outpatient utilization (C), and high overall utilization (D). Solid lines represent observed values, and dashed lines represent expected values. The x-axis represents quarters 1–12 over the 36-month period prior to PDAC diagnosis; the y-axis represents the proportion of patients with at least 1 health-care encounter (for any indication). ER, emergency room.
Table 1.
Characteristics of Pancreatic Ductal Adenocarcinoma Patients at the Time of Cancer Diagnosis, According to Their Pattern of Health-Care Utilization During the 36-Month Period Leading Up to Diagnosis, Indiana Network for Patient Care, Indianapolis, Indiana, 1999–2013
| Variable | Total (n = 1,023) | Pattern of Health-Care Utilization | P Value | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Late Acceleration (Group 1) (n = 430) | Early Acceleration (Group 2) (n = 99) | High Outpatient Utilization (Group 3) (n = 276) | High Overall Utilization (Group 4) (n = 218) | ||||||||
| No. of Patients | % | No. of Patients | % | No. of Patients | % | No. of Patients | % | No. of Patients | % | ||
| Age, yearsa | 67.4 (12.8) | 66.6 (12.6) | 70.2 (14.2) | 66.3 (11.8) | 68.9 (13.6) | 0.009 | |||||
| Female sex | 522 | 51.1 | 205 | 47.8 | 49 | 49.5 | 129 | 46.9 | 139 | 63.8 | <0.001 |
| Nonwhite race/ethnicity | 223 | 22.8 | 81 | 20.0 | 17 | 18.1 | 52 | 19.4 | 73 | 34.3 | <0.001 |
| Charlson comorbidity score at diagnosisa | 3.7 (2.6) | 3.2 (2.3) | 3.9 (2.3) | 3.5 (2.5) | 4.7 (2.8) | <0.001 | |||||
| Potential clinical biomarkers (any of the markers listed below)b | 555 | 54.3 | 171 | 39.8 | 64 | 64.6 | 156 | 56.5 | 164 | 75.2 | <0.001 |
| Diabetes mellitus | 313 | 30.6 | 63 | 14.7 | 38 | 38.4 | 96 | 34.8 | 116 | 53.2 | <0.001 |
| Diabetes mellitus diagnosed within 1.5 years of PDAC diagnosis | 97 | 9.5 | 41 | 9.5 | 13 | 13.1 | 28 | 10.1 | 15 | 6.9 | 0.334 |
| Any pancreatic disorder | 408 | 39.9 | 138 | 32.1 | 44 | 44.4 | 108 | 39.1 | 118 | 54.1 | <0.001 |
| Acute pancreatitis | 113 | 11.0 | 27 | 6.3 | 18 | 18.2 | 24 | 8.7 | 44 | 20.2 | <0.001 |
| Chronic pancreatitis | 50 | 4.9 | 17 | 4.0 | 7 | 7.1 | 7 | 2.5 | 19 | 8.7 | 0.008 |
| Other pancreatic disorder | 353 | 34.5 | 119 | 27.7 | 37 | 37.4 | 99 | 35.9 | 98 | 45.0 | <0.001 |
| Family history of GI tract neoplasms | 21 | 2.1 | 4 | 0.9 | 4 | 4.0 | 9 | 3.3 | 4 | 1.8 | 0.058 |
| PDAC tumor stagec | |||||||||||
| In situ | 5 | 1.0 | 3 | 1.2 | 0 | 0 | 2 | 1.8 | 0 | 0 | |
| Localized only | 88 | 17.5 | 37 | 14.6 | 16 | 25.4 | 17 | 15.3 | 18 | 24.0 | |
| Regional by direct extension only | 94 | 18.7 | 49 | 19.3 | 10 | 15.9 | 25 | 22.5 | 10 | 13.3 | |
| Regional lymph node involvement only | 38 | 7.6 | 18 | 7.1 | 6 | 9.5 | 6 | 5.4 | 8 | 10.7 | |
| Regional by both direct extension and lymph node involvement | 72 | 14.3 | 37 | 14.6 | 11 | 17.5 | 16 | 14.4 | 8 | 10.7 | |
| Regional, not otherwise specified | 1 | 0.2 | 1 | 0.4 | 0 | 0 | 0 | 0 | 0 | 0 | |
| Distant site(s)/node(s) involved | 153 | 30.4 | 87 | 34.3 | 10 | 15.9 | 34 | 30.6 | 22 | 29.3 | |
| Unknown whether there was extension or metastasis (unstaged, unknown, or unspecified) | 52 | 10.3 | 22 | 8.7 | 10 | 15.9 | 11 | 9.9 | 9 | 12.0 | |
Abbreviations: GI, gastrointestinal; PDAC, pancreatic ductal adenocarcinoma; SEER, Surveillance, Epidemiology, and End Results.
a Values are expressed as mean (standard deviation).
b A patient may have had 1 or more of these conditions, so percentages for the subgroups will not add up to 100%.
c Cancer stage was determined using SEER Program data linked to the Indiana Network for Patient Care.
Patient and encounter characteristics
Among 1,023 persons diagnosed with PDAC, 555 (54.3%) had 1 or more potential PDAC signals as defined in this study (Table 1). The most common signals included other pancreatic disorders (34.5%) and DM (30.6%). Of patients with DM who had 1 or more DM encounters during the 36-month period antecedent to PDAC diagnosis (n = 286), 282 (98.6%) had ICD-9 codes specific to type 2 DM. The prevalence of DM increased from group 1 (14.7%) to group 4 (53.2%) (P < 0.001). The same was true for any pancreatic disorder, including acute and chronic pancreatitis. In total, only 4.9% of cases had an antecedent diagnosis of chronic pancreatitis. There was no significant difference in the prevalence of recent-onset DM, defined as DM diagnosed within 1.5 years of the PDAC diagnosis, across groups (P = 0.334). Overall, recent-onset DM was observed in 9.5% of cases. Of those patients diagnosed with these PDAC signals, the majority of cases arose within 1 year of the cancer diagnosis, except for DM (Figure 2). Compared with any pancreatic disorders, which appeared a median of 0.04 years (interquartile range, 0.02–0.18) prior to the diagnosis of PDAC, DM appeared earlier (a median of 3.7 years prior (interquartile range, 1.1–7.1); P < 0.001).
Figure 2.
Cumulative proportion of pancreatic ductal adenocarcinoma (PDAC) patients with clinical signals and their time of onset, Indianapolis, Indiana, 1999–2013. “Any pancreatic disorder” represents a composite group comprising acute pancreatitis, chronic pancreatitis, and other pancreatic disorders.
Compared with late-acceleration individuals (group 1; 34.3%), fewer patients in the early-acceleration group (group 2; 15.9% (P = 0.006)) presented with metastatic PDAC. The frequency of metastatic PDAC at presentation was also lower for groups 3 (30.6%) and 4 (29.3%), but this was not statistically significant (for group 3 vs. group 1, P = 0.499; for group 4 vs. group 1, P = 0.427).
Pattern of related health-care encounters
The majority (75.7%) of patients had at least 1 health-care encounter for a related/potentially related diagnosis during the 36-month period leading up to the diagnosis of PDAC (Table 2); overall, the earliest related encounter occurred a median of 9.0 months (interquartile range, 0.9–28.2) before the cancer diagnosis. Patients in trajectory group 4 (high overall utilization) presented at a significantly earlier time point (30.3 months) relative to the cancer diagnosis compared with patients in group 1 (0.9 months; P < 0.001). Among those with a related antecedent encounter, patients in the early-acceleration (group 2; 13.9 months) and high outpatient utilization (group 3; 11.6 months) groups also presented significantly earlier. This pattern was most pronounced for outpatient encounters. Those in the early-acceleration, high outpatient, and high overall utilization groups were more likely to have at least 1 related encounter. The most common setting for these encounters was outpatient (61.9%), followed by inpatient (35.1%) and emergency room (21.0%). The indications for the earliest related encounter are summarized in Web Table 2; the most common related indications were abdominal pain (29.1%), DM (27.5%), back pain (14.9%), and weight loss (7.8%).
Table 2.
Patterns of Health-Care Utilization Among Pancreatic Ductal Adenocarcinoma Patients Prior to Cancer Diagnosis, Indiana Network for Patient Care, Indianapolis, Indiana, 1999–2013
| Type of Clinical Encounter and Variable | Total (n = 1,023) | Pattern of Health-Care Utilization | P Value | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Late Acceleration (Group 1) (n = 430) | Early Acceleration (Group 2) (n = 99) | High Outpatient Utilization (Group 3) (n = 276) | High Overall Utilization (Group 4) (n = 218) | ||||||||
| No. of Patients | % | No. of Patients | % | No. of Patients | % | No. of Patients | % | No. of Patients | % | ||
| Inpatient | |||||||||||
| Any encounter | 568 | 55.5 | 198 | 46.1 | 93 | 93.9 | 113 | 40.9 | 164 | 75.2 | <0.001 |
| Encounters with related diagnoses | 359 | 35.1 | 126 | 29.3 | 64 | 64.7 | 52 | 18.8 | 117 | 53.7 | <0.001 |
| Time between earliest encounter with a PDAC-related diagnosis and PDAC diagnosis, monthsa,b | 3.9 (0.7–19.2) | 1.3 (0.5–3.5) | 15.4 (3.5–26.6) | 1.9 (0.5–10.7) | 15.1 (3.5–26.0) | <0.001 | |||||
| Outpatient | |||||||||||
| Any encounter | 899 | 87.9 | 314 | 73.0 | 91 | 91.9 | 276 | 100.0 | 218 | 100.0 | <0.001 |
| Encounters with related diagnoses | 633 | 61.9 | 173 | 40.2 | 49 | 49.5 | 212 | 76.8 | 199 | 91.3 | <0.001 |
| Time between earliest encounter with a PDAC-related diagnosis and PDAC diagnosis, months | 9.5 (0.9–28.7) | 0.8 (0.3–2.5) | 5.6 (1.4–12.4) | 11.3 (1.6–26.1) | 30.2 (17.2–34.4) | <0.001 | |||||
| Emergency room | |||||||||||
| Any encounter | 579 | 56.6 | 184 | 42.8 | 90 | 90.9 | 126 | 45.7 | 179 | 82.1 | <0.001 |
| Encounters with related diagnoses | 215 | 21.0 | 42 | 9.8 | 33 | 33.3 | 48 | 17.4 | 92 | 42.2 | <0.001 |
| Time between earliest encounter with a PDAC-related diagnosis and PDAC diagnosis, months | 2.8 (0.5–17.2) | 0.8 (0.3–3.3) | 7.6 (2.3–21.8) | 0.7 (0.3–5.2) | 9.8 (0.9–22.9) | <0.001 | |||||
| Total | |||||||||||
| No. of encounters | 11 (4–26) | 3 (2–5) | 13 (9–23) | 17 (11–23) | 51 (32–72) | <0.001 | |||||
| Encounters with related diagnoses | 774 | 75.7 | 260 | 60.5 | 88 | 88.9 | 222 | 80.4 | 204 | 93.6 | <0.001 |
| Time between earliest encounter with a PDAC-related diagnosis and PDAC diagnosis, months | 9.0 (0.9–28.2) | 0.9 (0.4–3.3) | 13.9 (4.4–27.8) | 11.6 (1.6–26.2) | 30.3 (20.4–34.5) | <0.001 | |||||
Abbreviation: PDAC, pancreatic ductal adenocarcinoma.
a Values are expressed as median (interquartile range).
b Data are restricted to patients with at least 1 PDAC-related encounter.
Using trajectory group 1 (late-acceleration) patients as the reference category, group 4 patients were significantly older, more likely to be female and nonwhite, more likely to have higher baseline comorbidity, and more likely to have all clinical biomarkers (acute pancreatitis, DM, and other pancreatic disorders) except for chronic pancreatitis (Table 3). Patients with early acceleration (group 2) were significantly older and more likely to have DM and acute pancreatitis. Compared with group 1 patients, those with high outpatient utilization (group 3) could only be distinguished by higher odds of DM at the time of PDAC diagnosis.
Table 3.
Characteristics of Pancreatic Ductal Adenocarcinoma Patients Associated With Unique Prediagnosis Health-Care Utilization Trajectory Groups, Indiana Network for Patient Care, Indianapolis, Indiana, 1999–2013
| Variable | Odds Ratio | 95% Confidence Interval | P Value |
|---|---|---|---|
| Trajectory groupa 2 vs. group 1 | |||
| Age, years | 1.024 | 1.005, 1.044 | 0.013 |
| Female sex | 1.034 | 0.648, 1.648 | 0.89 |
| Nonwhite race/ethnicity | 0.729 | 0.400, 1.329 | 0.302 |
| Charlson comorbidity score | 1.069 | 0.973, 1.174 | 0.165 |
| Any potential clinical biomarker | |||
| Acute pancreatitis | 3.245 | 1.579, 6.665 | 0.001 |
| Chronic pancreatitis | 1.112 | 0.401, 3.084 | 0.839 |
| Other pancreatic disorder | 1.340 | 0.813, 2.208 | 0.251 |
| Diabetes mellitus | 3.127 | 1.851, 5.280 | <0.001 |
| Trajectory group 3 vs. group 1 | |||
| Age, years | 1.000 | 0.987, 1.013 | 0.979 |
| Female sex | 0.959 | 0.696, 1.320 | 0.795 |
| Nonwhite race/ethnicity | 0.837 | 0.560, 1.252 | 0.387 |
| Charlson comorbidity score | 0.993 | 0.925, 1.066 | 0.839 |
| Any potential clinical biomarker | |||
| Acute pancreatitis | 1.470 | 0.796, 2.712 | 0.218 |
| Chronic pancreatitis | 0.485 | 0.187, 1.258 | 0.137 |
| Other pancreatic disorder | 1.262 | 0.888, 1.794 | 0.195 |
| Diabetes mellitus | 3.036 | 2.060, 4.474 | <0.001 |
| Trajectory group 4 vs. group 1 | |||
| Age, years | 1.016 | 1.001, 1.031 | 0.035 |
| Female sex | 1.955 | 1.340, 2.852 | 0.001 |
| Nonwhite race/ethnicity | 1.563 | 1.030, 2.371 | 0.036 |
| Charlson comorbidity score | 1.159 | 1.079, 1.244 | <0.001 |
| Any potential clinical biomarker | |||
| Acute pancreatitis | 3.422 | 1.875, 6.244 | <0.001 |
| Chronic pancreatitis | 1.039 | 0.447, 2.416 | 0.929 |
| Other pancreatic disorder | 1.531 | 1.034, 2.268 | 0.034 |
| Diabetes mellitus | 4.738 | 3.146, 7.137 | <0.001 |
a Persons in trajectory group 1 (late acceleration, 42.0%) had minimal numbers of clinical encounters until the 2 quarters immediately prior to diagnosis of pancreatic ductal adenocarcinoma; group 2 individuals (early acceleration, 9.7%) had a pattern of increasing health-care utilization throughout the 36-month period; group 3 individuals (high outpatient utilization, 27.0%) had consistently high outpatient utilization but minimal inpatient/emergency room utilization; and group 4 individuals (high overall utilization, 21.3%) had consistently high utilization of all encounter types during the 36-month period.
DISCUSSION
Earlier detection of PDAC is a daunting task. This epidemiologic study illustrates that 42% of these individuals have minimal health-care encounters (i.e., late accelerators in terms of their health-care utilization) prior to their pancreatic cancer diagnosis. These patients are more likely to present in a clinical setting at an advanced stage. An additional 10% of individuals with PDAC identified in this cohort were excluded from the analysis due to their having no antecedent health-care encounters. In order to detect PDAC at an earlier stage in persons who have few or no clinical encounters prior to their presentation with PDAC-related symptoms, we will need to couple novel biomarkers with a unique public health effort to encourage patients to establish care with a primary care provider. The lifetime incidence of PDAC (1%–2%) would impede universal PDAC screening strategies akin to those used for breast and colon cancer, even if an excellent and minimally invasive screening test were available.
Clinical utility of clinical risk factors and harbingers of PDAC
Based on laboratory testing at the time of presentation with PDAC, a substantial number of patients with PDAC have DM at the time of clinical presentation (9); consistent with previous epidemiologic studies, this study highlights that far fewer patients (31%) are diagnosed with DM well in advance of the PDAC diagnosis (10). While there may be subclinical evidence of DM at the time of cancer presentation, this is unbeknownst to the patient in many cases. Since even fewer (10%) patients are diagnosed with DM within 1.5 years of PDAC, a screening program limited to recent-onset DM would not facilitate earlier detection in the majority of individuals with PDAC. Still, DM appears to be the most promising signal, since pancreatitis and other pancreatic disorders generally develop closer to the onset of PDAC itself. Similar to pancreatic cystic lesions, few patients with acute and chronic pancreatitis will progress to PDAC even when surveillance is initiated (10–12). However, the high baseline prevalence of DM in the US population (9%) (13) poses a challenge for identifying the small minority who will progress to PDAC. Patients in trajectory group 4 appear to be the best candidates for earlier diagnosis, despite their higher baseline comorbidity compared with other groups, given the higher baseline prevalence of PDAC signals (75%, including DM in 53%) and longer median time from the onset of related diagnoses (30 months). Although these represent a minority of all PDAC cases, future studies should explore the relative risk of developing PDAC among patients with the combination of DM and other PDAC risk factors, such as smoking, obesity, and family history of relevant cancers (11, 14, 15).
Only 9% of PDAC cases are localized at the time of diagnosis; this small group of patients has a 27% probability of 5-year survival due to the fact that their cancer is resected, as compared with 10% or less for patients with regional or metastatic disease, who do not qualify for resection (1). Perhaps a more realistic short-term goal for improving the prognosis of PDAC may be to diagnose the cancer at an earlier stage by recognizing its early signs, as suggested by Risch et al. (10), rather than implement preventive screening strategies. Since 42% of the cohort identified in the INPC entered the health-care system within months of their PDAC diagnosis, earlier detection and prevention remains a formidable challenge. Encouragingly, patients in groups 2, 3, and 4 had a potentially related health-care encounter early enough (medians of 13.9, 11.6, and 30.3 months, respectively) that if PDAC were sought out and diagnosed, their prognosis might have improved. Experts have suggested that shifting the diagnostic timeline by only 2 months could shift patients with stage III cancer at presentation to stage II (16). There is also a need to improve pancreatic cancer diagnostics, since endoscopic ultrasound-guided fine-needle aspiration has a sensitivity of only 80%–90% in expert hands (17).
Limitations
This study was limited by its retrospective design and the use of an electronic registry. To minimize detection bias, we limited the cohort to patients residing in the Indianapolis metropolitan area with a high (>90%) rate of electronic health data capture. Still, it is possible that some individuals sought care at non-INPC facilities earlier than at INPC facilities; this would not have applied to our trajectory groups but may have skewed the proportion of patients identified as “late acceleration” (group 1). Since we required 2 distinct PDAC-specific encounters for study inclusion (in an effort to maximize our specificity), it is plausible that patients with more advanced PDAC at presentation would have had only 1 PDAC encounter. It is unlikely that this would have changed our trajectory modeling, which focused on encounters that occurred prior to the PDAC diagnosis. Additionally, prevalences of other important PDAC risk factors, such as alcohol drinking, smoking, body mass index, and family history of relevant cancers, could not be reliably estimated from the INPC. In addition, the cohort did not include all patients diagnosed with DM and pancreatic disorders who never developed PDAC during the study period. Since prevalence statistics for these patients are not available within the registry, the relative odds of PDAC among these persons could not be estimated. However, this was not an objective of this study, and it is widely accepted that the prevalence of each of these factors far exceeds the incidence rate of PDAC; so the presence of one or another is insufficient to warrant PDAC screening (16).
Summary
While the cost burden of treating PDAC is well-studied (18–25), there is little epidemiologic evidence defining the patterns of health-care utilization leading up to the diagnosis of this cancer. The majority of patients diagnosed with PDAC have limited health-care encounters well in advance of their cancer diagnosis, posing a challenge for early-detection strategies. Few patients are diagnosed with DM, chronic pancreatitis, and other pancreatic disorders prior to presenting with PDAC. Still, there is an important minority of patients with high baseline utilization of the health-care system who appear to be most eligible for early-detection strategies. Ironically, while these patients have higher baseline comorbidity, including underlying DM and pancreatitis, more aggressive testing for PDAC may shift the diagnostic curve to earlier cancer stages or even cancer precursors.
Supplementary Material
ACKNOWLEDGMENTS
Author affiliations: Division of Gastroenterology and Hepatology, Medical University of South Carolina, Charleston, South Carolina (Gregory A. Cote); Department of Biostatistics, Richard M. Fairbanks School of Public Health, Indiana University–Purdue University Indianapolis, Indianapolis, Indiana (Huiping Xu); Division of Gastroenterology and Hepatology, School of Medicine, Indiana University, Indianapolis, Indiana (Jeffery J. Easler, Timothy D. Imler, Stuart Sherman); Regenstrief Institute, Indianapolis, Indiana (Timothy D. Imler, Evgenia Teal); Division of Endocrinology, School of Medicine, Indiana University, Indianapolis, Indiana (Murray Korc); Department of Biochemistry and Molecular Biology, School of Medicine, Indiana University, Indianapolis, Indiana (Murray Korc); and Pancreatic Cancer Signature Center at the Indiana University Simon Cancer Center, School of Medicine, Indiana University, Indianapolis, Indiana (Murray Korc).
This work was supported by a grant from the National Institutes of Health (National Institute of Diabetes and Digestive and Kidney Diseases grant 5K23DK095148) to G.A.C.
The views expressed in this article do not necessarily reflect the official policies of the National Institutes of Health. Each author certifies that he or she participated sufficiently in the work to believe in its overall validity and to take public responsibility for appropriate portions of its content.
Conflict of interest: none declared.
REFERENCES
- 1. Howlader N, Noone AM, Krapcho M, et al., eds. SEER Cancer Statistics Review, 1975–2012 Bethesda, MD: National Cancer Institute. http://seer.cancer.gov/csr/1975_2012/. Updated November 18, 2015. Accessed November 28, 2016.
- 2. Andersen DK, Andren-Sandberg A, Duell EJ, et al. Pancreatitis-diabetes-pancreatic cancer: summary of an NIDDK-NCI workshop. Pancreas. 2013;42(8):1227–1237. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3. McDonald CJ, Overhage JM, Barnes M, et al. The Indiana Network for Patient Care: a working local health information infrastructure. An example of a working infrastructure collaboration that links data from five health systems and hundreds of millions of entries. Health Aff (Millwood). 2005;24(5):1214–1220. [DOI] [PubMed] [Google Scholar]
- 4. Nagin DS, Tremblay RE. Analyzing developmental trajectories of distinct but related behaviors: a group-based method. Psychol Methods. 2001;6(1):18–34. [DOI] [PubMed] [Google Scholar]
- 5. Nagin DS. Analyzing developmental trajectories: semi-parametric, group-based approach. Psychol Methods. 1999;4:139–177. [DOI] [PubMed] [Google Scholar]
- 6. Jones BL, Nagin DS. Advances in group-based trajectory modeling and an SAS procedure for estimating them. Sociol Methods Res. 2007;35(4):542–571. [Google Scholar]
- 7. Jones BL, Nagin DS, Roeder K. A SAS procedure based on mixture models for estimating developmental trajectories. Sociol Methods Res. 2001;29(3):374–393. [Google Scholar]
- 8. Nylund KL, Asparouhov T, Muthen BO. Deciding on the number of classes in latent class analysis and growth mixture modeling: a Monte Carlo simulation study. Struct Equ Modeling. 2007;14(4):535–569. [Google Scholar]
- 9. Pannala R, Leirness JB, Bamlet WR, et al. Prevalence and clinical profile of pancreatic cancer-associated diabetes mellitus. Gastroenterology. 2008;134(4):981–987. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Risch HA, Yu H, Lu L, et al. Detectable symptomatology preceding the diagnosis of pancreatic cancer and absolute risk of pancreatic cancer diagnosis. Am J Epidemiol. 2015;182(1):26–34. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Alsamarrai A, Das SL, Windsor JA, et al. Factors that affect risk for pancreatic disease in the general population: a systematic review and meta-analysis of prospective cohort studies. Clin Gastroenterol Hepatol. 2014;12(10):1635.e5–1644.e5. [DOI] [PubMed] [Google Scholar]
- 12. Munigala S, Kanwal F, Xian H, et al. Increased risk of pancreatic adenocarcinoma after acute pancreatitis. Clin Gastroenterol Hepatol. 2014;12(7):1143.e1–1150.e1. [DOI] [PubMed] [Google Scholar]
- 13. National Center for Chronic Disease Prevention and Health Promotion, Centers for Disease Control and Prevention National Diabetes Statistics Report, 2014. Estimates of Diabetes and Its Burden in the United States Atlanta, GA: Centers for Disease Control and Prevention; 2014. http://www.thefdha.org/pdf/diabetes.pdf. Accessed September 1, 2016.
- 14. de Mutsert R, Sun Q, Willett WC, et al. Overweight in early adulthood, adult weight change, and risk of type 2 diabetes, cardiovascular diseases, and certain cancers in men: a cohort study. Am J Epidemiol. 2014;179(11):1353–1365. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. McWilliams RR, Maisonneuve P, Bamlet WR, et al. Risk factors for early-onset and very-early-onset pancreatic adenocarcinoma: a Pancreatic Cancer Case-Control Consortium (PanC4) analysis. Pancreas. 2016;45(2):311–316. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Chari ST, Kelly K, Hollingsworth MA, et al. Early detection of sporadic pancreatic cancer: summative review. Pancreas. 2015;44(5):693–712. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Brand RE, Adai AT, Centeno BA, et al. A microRNA-based test improves endoscopic ultrasound-guided cytologic diagnosis of pancreatic cancer. Clin Gastroenterol Hepatol. 2014;12(10):1717–1723. [DOI] [PubMed] [Google Scholar]
- 18. Du W, Touchette D, Vaitkevicius VK, et al. Cost analysis of pancreatic carcinoma treatment. Cancer. 2000;89(9):1917–1924. [DOI] [PubMed] [Google Scholar]
- 19. Hjelmgren J, Ceberg J, Persson U, et al. The cost of treating pancreatic cancer—a cohort study based on patients’ records from four hospitals in Sweden. Acta Oncol. 2003;42(3):218–226. [DOI] [PubMed] [Google Scholar]
- 20. Müller-Nordhorn J, Brüggenjürgen B, Böhmig M, et al. Direct and indirect costs in a prospective cohort of patients with pancreatic cancer. Aliment Pharmacol Ther. 2005;22(5):405–415. [DOI] [PubMed] [Google Scholar]
- 21. Chang S, Long SR, Kutikova L, et al. Burden of pancreatic cancer and disease progression: economic analysis in the US. Oncology. 2006;70(1):71–80. [DOI] [PubMed] [Google Scholar]
- 22. Abbott DE, Merkow RP, Cantor SB, et al. Cost-effectiveness of treatment strategies for pancreatic head adenocarcinoma and potential opportunities for improvement. Ann Surg Oncol. 2012;19(12):3659–3667. [DOI] [PubMed] [Google Scholar]
- 23. Ghatnekar O, Andersson R, Svensson M, et al. Modelling the benefits of early diagnosis of pancreatic cancer using a biomarker signature. Int J Cancer. 2013;133(10):2392–2397. [DOI] [PubMed] [Google Scholar]
- 24. Ljungman D, Hyltander A, Lundholm K. Cost-utility estimations of palliative care in patients with pancreatic adenocarcinoma: a retrospective analysis. World J Surg. 2013;37(8):1883–1891. [DOI] [PubMed] [Google Scholar]
- 25. DaCosta Byfield S, Nash Smyth E, Mytelka D, et al. Healthcare costs, treatment patterns, and resource utilization among pancreatic cancer patients in a managed care population. J Med Econom. 2013;16(12):1379–1386. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.