Use of Statewide Administrative Data to Assess Clinical Outcomes: A Retrospective Cohort Study of Therapeutic Anticoagulation for Isolated Calf Vein Thrombosis

Abstract

Background:

Single-center comparative effectiveness studies evaluating outcomes that can occur post-hospitalization may become biased if outcomes diagnosed at other facilities are not ascertained. Administrative datasets that link patients’ records across facilities may improve outcome ascertainment.

Objective:

To determine whether use of linked administrative data significantly augments thromboembolic outcome ascertainment

Research Design:

Retrospective cohort study

Subjects:

Patients with an acute isolated calf deep vein thrombosis (DVT) diagnosed at one Californian center during 2010-2013.

Measures:

Proximal DVT or pulmonary embolism (PE) within 180 days. We ascertained outcomes from linked California hospitalization, emergency department, and ambulatory surgery data and compared this information to outcomes previously identified from review of the center’s medical records.

Results:

Among 384 patients with an isolated calf DVT, 333 could be linked to longitudinal administrative data records. Ten patients had a possible proximal DVT or PE (4 more clearly so) from administrative data; all were unknown from medical record review. Eleven patients with known outcomes from medical record review had no outcome from administrative data. The adjusted OR of proximal DVT or PE with therapeutic anticoagulation attenuated from 0.33 (95% C.I. 0.12-0.87) using only medical record review to 0.64 (95% C.I. 0.29-1.40) using both medical record review and possible outcomes from administrative data. Restricting the outcome to diagnoses clearly involving proximal DVT or PE, the adjusted OR was 0.46 (95% C.I. 0.19-1.10).

Conclusions:

Use of linked hospital administrative data augmented detection of outcomes but imperfect linkage, non-specific diagnoses, and documentation/coding errors introduced uncertainty regarding the accuracy of outcome ascertainment.

INTRODUCTION

Thrombosis of the deep veins distal to the popliteal vein, or calf deep vein thrombosis (DVT), occurs commonly, involving an estimated 300,000 events/year in the U.S.¹ Although there is currently little consensus regarding optimal management of acute isolated calf DVT, a central question is whether therapeutic anticoagulation is warranted to reduce the likelihood of subsequent proximal DVT or pulmonary embolism (PE).^1–5

Administrative data involving hospitalizations and emergency department visits can provide longitudinal follow-up information about a known cohort of patients,⁶ including for studies involving venous thromboembolism. Potential impediments include confidentiality restrictions that discourage linking an individual’s records over time, limited specificity and misclassification of diagnoses, and lack of information from outpatient clinic visits. The validity of such administrative data for identifying DVT and PE has been good,^7–9 but coding changes introduced in 2009 particularly might help distinguish calf from proximal DVTs.

We recently conducted a single-center retrospective cohort study evaluating whether therapeutic anticoagulation might be associated with reduced subsequent proximal DVT or PE, as determined exclusively from medical records at our center. We sought to compare our original findings against an analysis incorporating linked statewide hospitalization, emergency department, and ambulatory surgery data to capture outcomes identified at other hospitals.

METHODS

Study Design and Population

We conducted a single-center retrospective cohort study, approved by our center’s Institutional Review Board. Use of California’s statewide datasets—provided by the California Office of Statewide Health Planning and Development—was approved by the state’s Committee for the Protection of Human Subjects.

We previously described details of the study population and methods.¹⁰ In brief, we identified all patients at our center age 18 years or older from January 1, 2010 through December 31, 2013 diagnosed on duplex ultrasound with an isolated calf DVT, defined as involving one or more of the deep veins distal to the popliteal vein—whether axial (posterior tibial, peroneal, tibioperoneal trunk) or muscular (gastrocnemius, soleal)—in the absence of a proximal DVT (common femoral, superficial femoral, deep femoral, or popliteal).

Exposure and Outcomes

We defined the exposure, therapeutic anticoagulation, as intention to administer therapeutic doses of unfractionated or low-molecular weight heparin (or analogous factor Xa inhibitor), warfarin, or a direct thrombin inhibitor. The primary outcome of the study—based on review of all information available in our center’s medical records—was radiographically confirmed proximal DVT or PE occurring within 180 days (including, potentially, outcomes detected at other hospitals described in our center’s records).

In this follow-up analysis, we also identified diagnoses potentially representing proximal DVT or PE (Table 1) subsequent to the index calf DVT from the California Patient Discharge, Emergency Department, and Ambulatory Surgery Datasets (hereafter, “statewide datasets”). We conducted an analysis of longitudinal statewide data compiled from 2007 through 2013 using a “record linkage number” (RLN) (a synthetic identifier based on social security number unique to each individual) to identify outcomes occurring at hospitals other than ours. We identified index records (revealing the appropriate RLNs) in the statewide datasets using deterministic matching of hospital identity, admission and discharge dates, age, and sex.

Table 1.

Diagnoses ascertained as potential proximal DVT or PE outcomes

Code	Description
415.11	Iatrogenic pulmonary embolism and infarction
415.13	Saddle embolus of pulmonary artery
415.19	Other pulmonary embolism
451.11	Phlebitis and thrombophlebitis of femoral vein (deep) (superficial)
451.19	Phlebitis and thrombophlebitis of other deep vessel of lower extremities
451.2	Phlebitis and thrombophlebitis of lower extremities, unspecified
451.81	Phlebitis and thrombophlebitis of iliac vein
451.9	Phlebitis and thrombophlebitis of unspecified site
453.2	Venous embolism and thrombosis of inferior vena cava
453.40	Acute venous embolism and thrombosis of unspecified deep vessels of lower extremity
453.41	Acute venous embolism and thrombosis of deep vessels of proximal lower extremity
453.50	Chronic venous embolism and thrombosis of unspecified deep vessels of lower extremity
453.51	Chronic venous embolism and thrombosis of deep vessels of proximal lower extremity
453.79	Chronic venous embolism and thrombosis of other specified veins
453.89	Acute venous embolism and thrombosis of other specified veins

Analysis

Patients in our study could have been diagnosed with the index calf DVT in an inpatient, emergency department, or outpatient setting at our center. We identified index records in the statewide datasets using known dates of prior admission/discharge or emergency department encounters at our center. For patients who had their index diagnosis during a non-surgical outpatient encounter (which would not be contained in the statewide datasets), we used dates from the chronologically nearest non-index inpatient, emergency department, or ambulatory surgery encounter, if available, to identify the appropriate RLN. We then used the RLN of the matched records in the statewide datasets to identify additional encounters at any hospital following the date of the index calf DVT diagnosis. We counted as outcomes proximal DVTs or PEs that occurred within 180 days after the diagnosis of the index calf DVT. For outcomes present on admission (POA) [POA status “Y” (diagnosis POA) or “W” (documentation is adequate but timing of diagnosis is unclear due to clinical uncertainty)], we considered the date of the encounter to represent the date of diagnosis. For outcomes not POA [POA status “N” (diagnosis not POA) or “U” (documentation is inadequate therefore timing of diagnosis cannot be determined)] occurring at another hospital, we assigned the date of diagnosis as the median day of that hospitalization.

We defined the primary outcome as either: (1) a diagnosis of a proximal DVT or PE from medical record abstraction at our center; or (2) a diagnosis code potentially representing proximal DVT or PE (Table 1) at another center from the statewide datasets, within 180 days after the index calf DVT. Among the outcome diagnosis codes (Table 1), 453.40 is not specified as either a proximal or distal DVT. Furthermore, a prior calf DVT—the inclusion criterion for participation in this study—might be captured using 453.40 instead of 453.42 (distal or calf). We therefore conducted sensitivity analyses considering: (1) a diagnosis of 453.40 alone as an outcome only if it was not POA; and (2) a diagnosis of 453.40 alone as not an outcome.

To help evaluate the reliability of outcome assessment from administrative data versus medical record review, we also examined the overlap between these two approaches for outcomes that occurred at our center, using medical record review as the criterion standard.

Because the records in the statewide datasets also included linked vital status records through 2011, we compared death within 180 days as a secondary study endpoint between the administrative data and our center’s records.

We linked records using SAS version 9.4 (SAS Institute Inc., Cary, NC) and conducted all subsequent analyses with Stata version 10 (StataCorp LP, College Station, TX) using two-tailed tests and an α level of 0.05.

RESULTS

Of 384 patients included in the original comparison, we identified the index record in the statewide datasets for 353, and 333 had an RLN identifier that allowed linkage to longitudinal records (Figure; also, see Supplemental Digital Content 1 for further explanation regarding unidentified records). Among the 353 patients with an identified index record, 89 (25%) had a subsequent encounter at the index hospital. Among the 333 patients with an RLN identifier, 18 (5%) had one or more encounters at a different hospital—regardless of whether an outcome diagnosis was coded—within 180 days after calf DVT diagnosis, encompassing 13 hospitalizations and 9 emergency department visits.

Figure.

Flow diagram of patients in study cohort

Administrative Data Outcomes at Other Hospitals

For 10 patients, encounters at another hospital were coded with a possible proximal DVT or PE within 180 days after the index calf DVT diagnosis, none of which were known to have occurred from review of our center’s records (Table 2). Four of the ten encounters involved one of the codes specific to proximal DVT or PE (415.19), and six involved 453.40 (three POA, three not POA) but not a code specific to proximal DVT or PE. No follow-up encounters were coded with any of the other potential proximal DVT or PE diagnoses.

Table 2.

Comparison of ascertainment of 180-day proximal DVT or PE outcomes between single-center medical record review and linked statewide administrative data among 384 patients with isolated calf DVTs

Outcome (proximal DVT or PE) information from linked statewide administrative data, n (%)a	Outcomes from medical record reviewb
	No proximal DVT or PE (N=363)	Proximal DVT only (N=11)	PE only (N=10)
No administrative data available (index record not identified)c	30 (8)	1 (9)	0 (0)
No code for proximal DVT or PE in administrative data from any hospital (the index hospital or another hospital)	312 (86)	8 (73)	3 (30)
Any code for proximal DVT or PE in administrative data from hospitals other than the index hospital	10 (3)d	0 (0)	0 (0)
415.19 POA	4 (1)	—	—
453.40 POA	3 (1)	—	—
453.40 not POA	3 (1)	—	—
Any code for proximal DVT or PE in administrative data only from the index hospital	12 (3)d	2 (18)	7 (70)
415.19 POA	2 (0.6)	0 (0)	5 (50)
415.19 not POA	0 (0)	0 (0)	2 (20)
453.40 POA	1 (0.3)	0 (0)	0 (0)
453.40 not POA	4 (1)	0 (0)	0 (0)
453.41 POA	3 (0.8)	1 (9)	0 (0)
453.41 not POA	2 (0.6)	1 (9)	0 (0)

DVT = deep vein thrombosis

PE = pulmonary embolism

POA = present on admission

^aAmong outcomes ascertained from the administrative data, we did not identify any of the codes from Table 1 that are not listed in this column.

^bWe did not identify any patients who had both a proximal DVT and a PE ascertained from medical record review.

^cThis row does not include the 20 patients for whom we identified the index administrative data record but who lacked an identifier to link to subsequent records (i.e., we could have ascertained an outcome for such patients from the administrative data only from the index encounter at the index hospital).

^dOne patient had a 415.19 code POA at a hospital other than the index hospital and a 453.40 code not POA at the index hospital, thus contributing to both the non-index hospital and index hospital rows in this Table

Combining the prior known outcomes from medical record review at our center (21 patients) with the administrative data outcomes at other hospitals (10 patients), 31 patients had an outcome (Table 3). The adjusted OR of proximal DVT or PE with therapeutic anticoagulation increased from 0.33 (95% C.I. 0.12-0.87) in the original analysis to 0.64 (95% C.I. 0.29-1.40) with the administrative data. This adjusted OR decreased with the more restrictive assumptions of sensitivity analyses (1) and (2). Considering any diagnosis codes potentially representing proximal DVT or PE as outcomes, the median (IQR) time to proximal DVT or PE was 18.5 (3, 54) days; considering only diagnosis codes specific to proximal DVT or PE as outcomes, it was 18.5 (8.5, 23) days (p=0.59, Mann-Whitney U test).

Table 3.

Therapeutic anticoagulation and proximal DVT or PE at 180 days among 384 patients with isolated calf DVTs

Source of outcome (proximal DVT or PE) information, n (%)	Control (N=141)	Therapeutic anticoagulation (N=243)	Unadjusted OR (95% C.I.)	Adjusted ORa (95% C.I.)
Medical record review at index hospital (original study results)	13 (9)	8 (3)	0.34 (0.14-0.83)	0.33 (0.12-0.87)
Administrative data from other hospitals
Any diagnosis potentially representing a proximal DVT or PEb	2 (1)	8 (3)
Diagnosis specific to proximal DVT or PEc	1 (1)	3 (1)
Diagnosis 453.40 not POA in the absence of a diagnosis specific to proximal DVT or PE	1 (1)	2 (1)
Diagnosis 453.40 POA in the absence of a diagnosis specific to proximal DVT or PE	0 (0)	3 (1)
Combination of medical record review at index hospital and administrative data from other hospitals (any diagnosis potentially representing a proximal DVT or PEb)	15 (11)	16 (6)	0.59 (0.28-1.24)	0.64 (0.29-1.40)
Index calf DVT included thrombosis of an axial veind	8/83 (10)	11/160 (7)	0.69 (0.27-1.79)	0.90 (0.32-2.52)
Index calf DVT included thrombosis of a muscular branch veind	10/83 (12)	8/132 (6)	0.47 (0.18-1.24)	0.37 (0.12-1.10)
Sensitivity analyses
(1) Diagnosis specific to proximal DVT or PEc or 453.40 not POA	15 (11)	13 (5)	0.47 (0.22-1.03)	0.53 (0.23-1.21)
(2) Diagnosis specific to proximal DVT or PEc	14 (10)	11 (4)	0.43 (0.19-0.98)	0.46 (0.19-1.10)

DVT = deep vein thrombosis

PE = pulmonary embolism

OR = odds ratio

C.I. = confidence interval

POA = present on admission

^aAs per the original study, adjusted for age, sex, care setting at the time of calf DVT diagnosis, existing cancer, and history of prior DVT or PE

^b“Any diagnosis potentially representing a proximal DVT or PE” is defined as diagnosis codes 415.11, 415.13, 415.19, 453.40, or 453.41, independent of POA status

^cA “diagnosis specific to proximal DVT or PE” is defined as diagnosis codes 415.11, 415.13, 415.19, or 453.41, independent of POA status

^dAxial and muscular branch categories are not mutually exclusive, so both categories might apply to the same patient.

One PE outcome known from our center’s records to have been diagnosed at another hospital was not apparent in the administrative data. (The index encounter was identified but it did not have a RLN.)

Administrative Data Outcomes at Index Hospital

Among 353 patients for whom we identified the administrative data index record, 21 had a code potentially representing a proximal DVT or PE at our center within 180 days, but only 9 of these had a verified outcome from medical record review (i.e., there were 12 false positives) (Table 2). One of the false-positive cases involved only 453.40 POA, and an additional four false-positives involved only 453.40 not POA. None of the nine true-positive cases involved only 453.40 (whether POA or not). Of the 12 false-positive cases, 8 involved erroneous or suboptimal coding (4 involved use of 453.40 when 453.42 would have been more appropriate and 4 involved an incorrect code), 2 involved physician documentation errors, and 2 involved clinical diagnoses that were not radiographically evaluated (for study purposes, we defined the outcome to require radiographic confirmation).

Eleven patients without a code potentially representing a proximal DVT or PE had a verified outcome from medical record review (false negatives).

Deaths

From the administrative data, we identified three additional patients (one anticoagulated, two not anticoagulated) who died within 180 days of diagnosis of the calf DVT. Without this information, we estimated the adjusted OR of death as 0.73 (95% C.I. 0.34-1.56), and with the information, 0.67 (95% C.I. 0.32-1.42). A composite outcome of proximal DVT, PE, or death based on the administrative data (considering only diagnosis codes specific to proximal DVT or PE as outcomes) changed from an adjusted OR of 0.46 (95% C.I. 0.24-0.88) to 0.49 (95% C.I. 0.26-0.91).

DISCUSSION

Although administrative data augmented outcome assessment for this cohort study, these data also raised substantial problems in the ascertainment of proximal DVT outcomes. Compared with our original findings, inclusion of outcomes from administrative data suggests therapeutic anticoagulation is not clearly associated with reduced likelihood of subsequent proximal DVT or PE (albeit a type II error is possible). The non-specific 453.40 diagnosis code failed to distinguish between proximal and distal lower extremity DVTs, introducing uncertainty about the reliability of administrative data-based outcomes. Additionally, administrative data missed several true outcomes.

Coders are instructed to apply the most specific code possible supported by physician documentation and other records.¹¹ However, if the documentation is vague or coders do not recognize clues specifying the location of DVTs (e.g., ultrasound reports), coders will favor non-specific diagnoses like 453.40 over specific ones like 453.41 or 453.42. The anatomic specificity and increasingly widespread use of ultrasound suggests that 453.40 may be an unnecessary category for classifying lower extremity DVTs, but the National Center for Health Statistics and coding organizations have favored retaining non-specific options in the coding classification. The implementation of ICD-10-CM did not resolve this issue because the non-specific I82.40 category of codes (“Acute embolism and thrombosis of unspecified deep veins of lower extremity”) similarly applies to both proximal and distal veins.

Prior studies have shown a relatively high positive predictive value of PE- and DVT-related codes^{8, 9} and moderate reliability of POA status for these codes.⁷ However, based on our limited analysis, the coding may be less accurate particularly in distinguishing proximal from distal DVT, and the non-specific DVT code (453.40) alone was more clearly associated with false positives.

Because the 453.40 diagnosis code proved erroneous as an outcome among administrative data from our own center (verified against medical record review), we feel that sensitivity analysis (2) provides the most reliable summary finding of our study. Considering the limitations of both our original study and the administrative data we analyzed here, we conclude that therapeutic anticoagulation for isolated calf DVTs may be a reasonable option, but—with a confidence interval for the adjusted OR that includes 1—it is not clearly beneficial.

More generally, the use of administrative data for outcome ascertainment has the potential to improve the validity of comparative effectiveness studies, but the applicability of these data to the question at hand should be carefully evaluated and their reliability should be independently verified if possible. Specifically, investigators should consider that attrition of subjects from imperfect linkage, non-specificity of diagnoses, and documentation and coding errors can all contribute to suboptimal outcome ascertainment.