Step 1: Identify definite PAD patients



Prologue

by Mark Arratoon

The next few pages are a prepend to the PAD use case downloaded from the Project 3 web site (). This prepend is completely independent of the original use case and just intended to illustrate the potential role of process-driven design in algorithm implementation.

This introduction should also be considered within the larger context of architecture implementations well-used in the IT industry as a whole for many years now. Specifically, the central role an Enterprise Service Bus (ESB) has assumed in modern message-driven, SOA infrastructures, e.g., NHIN CONNECT itself is built on top of Oracle/Sun’s GlassFish ESB.

For more on ESBs, you can start at:



Therein is a table of major functions provided or supported directly or indirectly by an ESB:

|Category |Functions |

|Invocation |support for synchronous and asynchronous transport protocols, service mapping (locating and binding) |

|Routing |addressability, static/deterministic routing, content-based routing, rules-based routing, policy-based routing |

|Mediation |adapters, protocol transformation, service mapping |

|Messaging |message-processing, message transformation and message enhancement |

|Process choreography[3] |implementation of complex business processes |

|Service orchestration² |coordination of multiple implementation services exposed as a single, aggregate service |

|Complex event processing |event-interpretation, correlation, pattern-matching |

|Other quality of service |security (encryption and signing), reliable delivery, transaction management |

|Management |monitoring, audit, logging, metering, admin console, BAM |

This prepend is focused on exploiting process choreography/service orchestration using a standard process execution language (BPEL). Such process-driven flows are typically used in combination with, or in place of, simpler routing/filtering/transformation-based flows, all being deployed on an ESB-based infrastructure.

Another open source ESB is Mule ESB, the underlying technology Mirth Connect is built on which adds it’s own healthcare-specific features and customizations (Mirth Connect is currently being considered by the infrastructure project for adoption - ).

Example ESB architecture pictures are shown below, one from Mule ESB, the other from Oracle (note different products have different constructs encapsulating ESB core routing/filtering/tranformation capabilities, eg., Mirth uses “channels”, Apache Camel uses “routes”, Oracle uses “mediators”).

Mule ESB (open source):

[pic]

Oracle SOA Suite (commercial offering):

[pic]

Below is a list of some (not all) BPEL engines. Apache ODE (Orchestration Director Engine) donated several years ago by Intalio, is the most popular open source choice, also powering other offerings such as Intalio’s BPM product line, WSO2’s Business Process Server and JBoss’s Riftsaw.

|Product |Product Page|Vendor |Edition |Release Date |Framework |Compatibility |License |

|ActiveVOS |[1] |Active Endpoints [2] |8.0 |September 2010 |Servlet or Java |BPMN 2.0; WS-BPEL; |Proprietary |

| | | | | |EE |BPEL4People / WS-HumanTask; | |

| | | | | | |standards | |

|Apache ODE |[3] |ASF |1.3.4 |9 June 2010 |Apache Axis, JBI |BPEL4WS 1.1, WS-BPEL 2.0 |Apache |

| |[4] |(donated by Intalio) |1.0.164 |7 June 2006 |Java EE |( WS-HumanTask with Apache | |

| | | | | | |HISE ) | |

|BizTalk Server |[5] |Microsoft |Biztalk |2010 |.NET |BPEL, BPMN, RFID, WSDL, ... |Proprietary |

| | | |2010 | | | | |

|iBolt Server |[6] |Magic Software | | |Java EE |BPEL4WS |Proprietary |

| | |Enterprises | | | | | |

|jBPM |[7] |jBoss |3.3.1 |12 January 2009 |Java EE |WS-BPEL |LGPL |

|Open ESB |[8] |Sun Microsystems |2.0 |10 February 2009|Java EE, JBI |WS-BPEL 2.0 |Open Source, CDDL|

|Oracle BPEL Process |[9] |Oracle Corporation |11g |April 2010 |Java EE |WS-BPEL 2.0, BPMN |Proprietary |

|Manager | | | | | | | |

|(formerly Collaxa BPEL | | | | | | | |

|Orchestration Server) | | | | | | | |

|OW2 Orchestra |[10] |OW2 |4.6 |28 October 2009 |Apache Axis |WS-BPEL 2.0 |LGPL |

| | | | | |Apache CXF | | |

| | | | | |OSGi | | |

| | | | | |Java EE | | |

|Parasoft BPEL Maestro |[11] |Parasoft |5.0.1 |25 February 2010|Servlet |WS-BPEL, BPEL4People / |Proprietary |

| | | | | | |WS-HumanTask | |

|Petals BPEL Engine |[12] |Petals Link |1.0.1 |08 December 2009|Java EE |WS-BPEL 2.0, WSDL 1.1 and 2.0|LGPL |

| |[13] | | | | | | |

|SAP Exchange |[14] |SAP AG |3.0 | | |BPEL |Proprietary |

|Infrastructure | | | | | | | |

|Virtuoso Universal Server|[15] |OpenLink Software |4.5 |2006 | |UDDI, WS-BPEL, WS-* |GPL and |

| | | | | | | |Proprietary |

|WebSphere Process Server |[16] |IBM |6.0.1.3 |29 September |Java EE |WS-BPEL |Proprietary |

| | | | |2006 | | | |

PAD Process Representation #1 : The PAD algorithm as a sequential flow thru all sections

[pic]

An exploded version of the above scoped section flow is shown below.

Notes:

• As we proceed from Section 2-5, the # of matches are accumulated. A final tally and PAD assessment is made prior to return of definite, probable or negative assessment for PAD.

• The flow presupposes existing services (“Partner Links”) for extracting EHR data such as codes and medications.

• Such services can be either synchronous or asynchronous in nature and via ESB adaptive technology need not be SOAP-based. Human-based tasking services can also be part of the mix.

• Canonical data transformations may occur either via the ESB’s native capabilities or via specific transformation services provisioned on the bus.

• The boxes that end in ‘Rules’ (see exploded version below) represent either hard-coded process-specific rules (e.g., expressed in Java, XPath, SQL) OR more flexibly, invocation of an ESB provisioned rule engine. Rules can then be dynamically re/configured and done so independent of any specific process flow.

[pic]

PAD Process Representation #2: The PAD algorithm with early exit on a “definite” assessment [pic]

This is an alternate flow in which we decide to exit the process on immediate determination of a definite PAD assessment. All other internal processing within the grey scoped sections is reused from representation #1 above.

The above Proof-of-Concept examples just demonstrate the agility a process-based design can bring. Algorithms can be adjusted as new evidence-based criteria is discovered via changes to rules or process flow design without affecting other components (i.e., services, data access layers, etc.).

End of Prologue

Peripheral Arterial Disease (PAD)

Peripheral Arterial Disease (PAD) is prevalent with approximately 10-12 million adults in US affected. For those with PAD, morbidity and mortality are high and quality of life is markedly impaired. The genetic basis of PAD is poorly understood and is the focus of the Mayo Clinic Electronic Medical Records and Genomics (eMERGE) Network study.

In this document, we provide pseudocode modeling a standardized diagnostic algorithm for PAD. A set of ‘gold-standard’ rules were developed for the Mayo Clinic population as well as billing rules for transportability to sites without vascular laboratory data. Depending on the values of relevant variables, the rules classify a patient into one of the 2 classes: presence of PAD, and absence of PAD.

Ankle-brachial index (ABI) for ascertaining PAD

The ankle-brachial index (ABI) was measured in Mayo Clinic’s non-invasive vascular laboratory in Rochester MN using a standardized protocol. Following a 5-min rest, subjects were evaluated in the supine position. Appropriately-sized BP cuffs were placed on each arm and ankle, and a Doppler ultrasonic instrument (Medisonics, Minneapolis MN) was used to detect arterial signals. The cuff was inflated to 10 mm Hg above SBP and deflated at 2 mm Hg/s. The first reappearance of the arterial signal was taken as the SBP. To calculate the ABI, the SBP at each ankle site (posterior tibial and dorsalis pedis arteries) was divided by the higher of the two brachial pressures. The lower of the average ABIs from the two legs was used to define PAD. An ABI cut-point of less than or equal to 0.90 was used to diagnose PAD. Subjects with poorly compressible arteries due to medial arterial calcification were also classified as having PAD. The criteria for poorly compressible arteries in the Mayo vascular lab are as follows: an ABI >1.4, an ankle BP >255 mmHg, or having non-reproducible ankle BP measurements.

EMR Extraction of ABI Phenotype

To define a ‘definite’ PAD case population, patients will:

meet the algorithmic vascular lab criteria in Section 1 or

have any of the two related codes or presence of indicative phrases in Sections 2-5.

To define a ‘probable’ PAD case population for further investigation, patients will beet

any one of the related codes or presence of indicative phrases in Sections 2-5.

Section 1: Vascular Lab Criteria

Ankle brachial index 255 mmHg, or any ABI >1.4

Exclude non-atherosclerotic causes of PAD:

≥2 occurrences of the following codes

747.22, 237.7, 443.1, 446.0, 446.4, 446.5, 446.6, 446.7, 447.6, 710.1, 747.1, 747.64.

Section 2: ICD-9-CM Diagnosis Codes For PAD

440.2×, 440.3×, or 440.8×.

Section 3: Procedure Codes Related To PAD

One of the ICD-9-CM/CPT-4 procedure codes for lower extremity artery angiography: 88.48, 75710, 75711, 75712, 75716, 75717, 75718, 75630, 75631. PLUS one (concurrent) of the codes below for non coronary vessel stents: 39.50, 39.90, 37205, 37206, 37207, 37208, 37184, 37185, 37186.

OR

One of the ICD-9-CM/CPT-4 procedure codes for lower extremity artery surgical and percutaneous vascular interventions: 38.18, 39.50, 39.25, 39.29, 38.08, 38.38, 38.48, 39.49; 39.56, 39.57, 39.58, 39.90, 35302, 35303, 35304, 35305, 35306, 35331, 35351, 35355, 35361, 35363, 35371, 35372, 35381, 35452, 35454, 35456, 35459, 35470, 35472, 35473, 35474, 35481, 35482, 35483, 35485, 35491, 35492, 35493, 35495, 35521, 35533, 35537, 35538, 35539, 35540, 35541, 35546, 35548, 35549, 35551, 35556, 35558, 35563, 35565, 35566, 35571, 35582, 35583, 35585, 35587, 35621, 35623, 35637, 35638, 35641, 35646, 35647, 35651, 35654, 35656, 35661, 35663, 35665, 35666, 35671, 35226, 35256, 35286, 35700, 35721, 35741, 35876, 35879, 35881, 35883, 35884, 37184, 37185, 37186, 37205, 37206, 37207, 37208.

Exclude if one of the following ICD-9-CM codes for alternate reasons for surgery is also present: 736.3×, 736.4×, 736.5, 736.6, 736.7×, 736.8×, 736.9, 735.×, 754.3×, 754.4×, 754.5×, 754.6×, 754.7×, 755.02, 755.13, 755.14, 755.3, 755.4, 755.6×, 755.8, 759.7, 759.89, 895.××, 896.××, 897.××, 820.××, 821.××, 822.××, 823.××, 824.××, 825.××, 826.××, 827.××, 828.××, 829.××, 835.××, 836. ××, 837.××, 838.××, 904.××, 928.××, 929.××, 959.6, 959.7, 996.4×, 996.66, 996.67, 996.77, 996.78.

OR

One of the ICD-9-CM/CPT-4 procedure codes for lower extremity amputation: 84.1×, 84.91, 27295, 27590, 27591, 27592, 27598, 27880, 27781, 27782, 27888, 27889, 28800, 28805.

Exclude if one of the following ICD-9-CM codes for non-vascular amputation is also present: 170.6, 170.7, 170.8, 170.9, 171.3, 172.7, 173.7, 198.5, 344.1, 711.0, 728.86, 733.2, 736.3×, 736.4×, 736.5, 736.6, 736.7×, 736.8×, 736.9, 735. ×, 754.3×,754.4×,754.5×,754.6×,754.7×,755.02, 755.13, 755.14, 755.3, 755.4, 755.6×, 755.8, 759.7, 759.89, 820.××, 821.××, 822.××, 823.××, 824.××, 825.××, 826.××, 827.××, 828.××, 829.××, 835.××, 836. ××, 837.××, 838.××, 890. ××, 891, 895.××, 896.××, 897.××, 904.××, 905.4, 928.××, 929.××,959.6, 959.7,996.4×, 996.66, 996.67, 996.77, 996.78.

Mutivariable logistic regression model to ascertain PAD.

This model was based on billing codes and analysis of the vascular test (ABI) results in the vascular lab database at Mayo Rochester. In this database of 22156 individuals studied over the last decade, 14408 had PAD and 7748 did not.

This model uses the following billing code variables to predict presence of PAD in these patients.

|Variable |DF |Estimate |StdErr |WaldChiSq |ProbChiSq |Score |

|Intercept |1 |-1.58 |0.04 |1421.53 |0.0000 |-6 |

|CPT4Px73725 |1 |0.66 |0.16 |16.33 |0.0001 |3 |

|CPT4Px75635 |1 |0.54 |0.11 |24.09 |0.0000 |2 |

|CPT4Px75710 |1 |0.61 |0.13 |21.55 |0.0000 |2 |

|CPT4Px75716 |1 |0.75 |0.12 |36.69 |0.0000 |3 |

|CPT4Px93922 |1 |0.53 |0.04 |159.73 |0.0000 |2 |

|ICD9Px8848 |1 |0.65 |0.13 |23.94 |0.0000 |3 |

|ICD9Dx44020 |1 |1.25 |0.05 |708.82 |0.0000 |5 |

|ICD9Dx44021 |1 |2.16 |0.05 |2039.31 |0.0000 |9 |

|ICD9Dx44022 |1 |1.04 |0.11 |83.75 |0.0000 |4 |

|ICD9Dx44023 |1 |1.62 |0.07 |567.05 |0.0000 |6 |

|ICD9Dx44024 |1 |1.91 |0.18 |106.94 |0.0000 |8 |

|ICD9Dx4408 |1 |0.74 |0.16 |21.91 |0.0000 |3 |

|ICD9Dx4409 |1 |0.93 |0.04 |432.30 |0.0000 |4 |

|ICD9Dx4439 |1 |0.84 |0.04 |462.36 |0.0000 |3 |

An integer score based on the beta-coefficients was created and used to predict presence of PAD.

The results of our study are shown in the ROC curve below. The AUROC was 0.86. A score of 8 had ~80% sensitivity and specificity for the prediction of PAD.

A description of the billing codes is provided in the table below.

|Code |Estimate |Score |Code Description |

|CPT4Px73725 |0.66 |3 |MRA Lower extremity w/ or w/o contrast |

|CPT4Px75635 |0.54 |2 |CT angiogram - abdominal aorta and bilateral iliofemoral lower extremity runoff |

|CPT4Px75710 |0.61 |2 |Angiography, extremity, unilateral, radiological supervision and interpretation |

|CPT4Px75716 |0.75 |3 |bilateral angiograph extremity |

|CPT4Px93922 |0.53 |2 |Non-invasive physiologic studies of lower extremity arteries, single level, bilateral |

|ICD9Px8848 |0.65 |3 |Arteriography of femoral and other lower extremity arteries |

|ICD9Dx44020 |1.25 |5 |Atherosclerosis of native arteries of the extremities, unspecified |

|ICD9Dx44021 |2.16 |9 |Atherosclerosis of native arteries of the extremities with intermittent claudication |

|ICD9Dx44022 |1.04 |4 |Atherosclerosis of native arteries of the extremities with rest pain |

|ICD9Dx44023 |1.62 |6 |Atherosclerosis of native arteries of the extremities with ulceration |

|ICD9Dx44024 |1.91 |8 |Atherosclerosis of native arteries of the extremities with gangrene |

|ICD9Dx4408 |0.74 |3 |Atherosclerosis of other specified arteries |

|ICD9Dx4409 |0.93 |4 |generalized and unspecified atherosclerosis |

|ICD9Dx4439 |0.84 |3 |Peripheral vascular disease, unspecified |

Section 4: NLP Of Radiology Reports

Assess presence of phrases indicative of lower limb artery stenosis/occlusion in one or more sentences through Natural Language Processing (NLP) annotation techniques.

Key words for lower extremity arteries: left leg; right leg; left lower extremity; right lower extremity; bilaterally; Iliac artery (arteries); femoral artery (arteries); SFA; Superficial (arteries); Profunda (arteries); Deep femoral (arteries); Popliteal (arteries); tibioperoneal trunk (trunks); peroneal (arteries); Anterior tibial (arteries); AT; Posterior tibial (arteries); PT; runoff; femoral- femoral (arteries); femoral-to-peroneal artery; fem-pop; fem-fem; calf artery (arteries); common femoral; external iliac; common iliac artery; internal iliac arteries; iliac systems; origin of iliac; CFA; femoropopliteal artery; ileofemoral; aortobifemoral; axillobifemoral; aortobiiliac; aortobi-iliac

Key words for occlusive arterial disease: extensive collateral formation; collateral; occluded; occlusion; occludes; proximal occlusion; complete occlusion; one-vessel runoff; one-vessel run-off; single vessel runoff; single vessel run-off; high-grade stenosis; high grade stenosis; severe atherosclerotic stenosis; significant stenosis; moderate stenosis; moderate focal stenosis; multi-focus stenosis; moderate focal stenosis; severe narrowing; moderate narrowing; moderate atheromatous changes; moderate atheromatous plaque; not demonstrate flow; significant atheromatous plaque; severe plaque; significant atheromatous; moderate atheromatous changes; severe atherosclerotic disease; ASO; PTA; angioplasty; recanalization; endarterectomy

Section 5: Use Of Medications For Claudication:

Assess presence of medications for Claudication.

Cilostozol (Pletal) or Pentoxifylline (Trental, Pentoxil, Pentopak).

-----------------------

[pic]

[pic]

................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download