William R. Dailey Statement of Work for NLP/CAC 10/05/2014
Document Control Sheet General Information Project Name Project Manager Business Owner (Key Sponsor) Provider Single Point of Contact Document Preparation Information Author Date Organization Name Phone Number E-Mail File Location (link) Distribution and Approvals Name Title and Organization Signature Approval Date Change History Date Change Description Approved By
1. Background Golden Valley Memorial Healthcare (GVMH) is a healthcare system comprised of one hospital and a multispecialty clinic with 4 clinic locations. The multispecialty clinic has 30 providers, 6 of which are specialists and 24 are primary care providers. Inpatient volumes continue to decrease at approximately 4-5% per year while outpatient volumes continue to grow. Over the past 5 years, the healthcare system profit margin has been driven by this increased outpatient revenue. Optimizing efficiency with regard to this revenue stream is critical to the continued success of GVMH. Increased pressures on coding accuracy coupled with this increase in volume have created a large opportunity for improvement in this revenue stream. There are significant bottle-necks in the processes that currently exist between the point of care and the resulting reimbursement. The methods previously in place for coding, billing and subsequent reimbursement have not scaled well in response to increased outpatient volumes. It is helpful to review the current processes for a basis in understanding the impact of potential solutions. Status Quo The current system can be broken down into several base components that are important with regard to optimization. These components begin at the patient encounter with the electronic medical record (EMR), traverse coding and billing prior to submission to the payer for reimbursement. Descriptions of each of these base components are outlined below.
EMR The electronic medical record system used in the outpatient setting is McKesson IC-Chart (formerly Med-3000/InteGreat). This system allows the providers to document both discrete and free text items either concurrently or subsequent to the encounter. EMR functionality and individual provider preference and efficiency-thoroughness tradeoff dictate the variation from provider to provider regarding granular versus free text data entry for each encounter. All fixed historical data are fully granular within the EMR. Free text entries are classified in the typical fashion of the SOAP format within the note and as such are actually classified free text. Examples of this would be a free text blocks classified as History of Present Illness, Review of Systems broken down by system, Exam broken down by region or specialty, Diagnosis and Plan. Each of these can be discretely selected and coded by MEDCIN ID also via radio buttons and selection trees. Generally, each progress note contains a blend of discrete items and these classified free text items, again efficiency determines this blend. Once documentation on the encounter is complete, the provider finalizes the note within the EMR and an E&M code (level) is automatically determined based on the discrete data alone. The provider has the opportunity at this time to change that level to a more appropriate level based on what they have done within the free text portions of the note. Some providers do this well and some leave the code as selected based on the discrete data alone resulting in under-coding. The note is then approved and is forwarded to coding. The average note is completed and forwarded within 2 days with the vast majority completed concurrently with the patient encounter.
Coding The next step is for coding to manually extract and assign ICD-9/10 codes for all free text diagnoses and do all required compliance checks to ensure each encounter is appropriately coded and has the proper E&M level as supported by the entirety of the documentation (discrete and free text). This is a computer assisted (Encoder Pro) manual process that has resulted in an increasing backlog within the coding department as volumes have increased. The current average backlog within coding is 21 days. Once this is complete the appropriately coded and checked super-bill is forwarded to the business office for billing. Billing The business office receives the super-bill, verifies payer information and submits the claim to the appropriate payer. There are other business processes involved resulting in a 10 days average delay in charge submission. Payer Various payers have different turn around times for reimbursement. These reimbursement times vary from 14 days for private payers, to 30 days for Medicaid and Medicare. The average time to reimbursement is approximately 24 days. This results in average account receivable (AR) days in billing of approximately 34. Cost Each day of delay in these processes has a cost associated with it. Figure 1 shows the overall process and associated cost.
Figure 1. Process and Cost Figure 1 demonstrates the process starting at the date of Service (DOS) and progressing through reimbursement ($). Discussions with the clinic administrator reveal the annual cost per average day delay attributed to each portion of the process. As and example if average AR days are shortened by 1 day this would results in an increased income of $100,000 over the entire years. These numbers are determined based on the volume, E&M/CPT level and payer-mix specific to GVMH. There is a nominal cost to recoding that is calculated via the contribution to average AR days. This <5% rejected result in an additional 7 day delay which contributes approximately 0.35 days to the average AR days. The bottom line is that there is an annual 2.3 million dollar potentially avoidable cost associated with delays in the outpatient coding and billing cycle at GVMH. The payer delay is essentially fixed (although there may be some short-cycle potential if
direct to bill were utilized). This is a significant cost and there is great interest in increasing the efficiency of this process. 2. Objectives The overarching objective of this project would be to increase coding and billing efficiency to shorten the time between DOS and reimbursement to recoup a portion of the cost outlined. The calculations are straightforward here. Each day saved in the coding process results in $65,000 savings over the year. Similarly, each day saved in billing results in $100,000 savings over the year. The overall cost being $2.3 Million, the goal is to reduce the cost by 10% resulting in an increase in revenue of $230,000 annually. This objective could be accomplished using Natural Language Processing (NLP) and Computer Assisted Coding (CAC). NLP technology is quite new as applied to medicine but has been around for quite some time in other applications (3M). Figure 2 shows the basic concept (3M). Figure 2. NLP and CAC overview
The basics are that the free text entries are read and mapped into the appropriate medical ontologies (SNOMED-CT, ICD-9/10, RxNorm). These are then combined with the already discrete data captured in the balance of the note creating a fully-discrete version of the note that can then be fed into the encoder. The staff that do the coding then become more editors than coders increasing their productivity. Productivity increases from 20-89% have been realized in outpatient settings (Optum, Nuance). Minimally, NLP and CAC should be capable of delivering on at least the low end of that spectrum (20%). Applied to this case that would allow for a potential shortening of the current 21 day coding delay to around 19 days. It is likely that some direct to billing would also occur over time, reducing some of the time in billing (conservative estimate of a 1 day equivalent). Additional reductions would come in the form of eliminating or reducing the needs for recoding (0.35 days). The bottom line annual savings would therefore be: 2 x $65,000 = $130,000 1.35 x $100,000 = $135,000 Total Annual Savings estimate: $265,000 Using the lowest quoted improvements in efficiency is very conservative and compares favorably to the overall goal of a $230,000 savings annually. The actual efficiency and improvements have the potential to be much higher. Even using a 40% improvement would double this result and that is still half of quoted maximum improvements (89% Jamaica Nuance).
Constraints Cost the initial cost of the system should be set at a breakeven for year 1. This results in an initial budget of $230,000 for purchase/implementation. Implementation interfaces, hardware and training would have to be included in that cost. Ongoing - The ongoing support and licensing should be less than the differential between goal and minimum anticipated ($35k) but typical licensing would place it at 15% or < $23,000 annually. 3. Scope Evaluate vendor solutions meeting business and budgetary requirements within 2 months. Get acceptance guarantee regarding efficiency from vendor as part of selection (20%) Complete acquisition via contract within 6 months. Implement solution within 3 months of executing contract (interfaces, installation, training). Evaluate efficiency monthly as an in-process measure. Seek solutions/remedy from vendor on an ongoing basis for 1 year (approaching or exceeding 20%). Evaluate outcome efficiency at the end of year 1 (> 20%).
4. Deliverables Deliverable Responsible Criteria Due Approver RFI Selection Group Within 1 month 10/5/14 CMIO RFP CMIO Within 2 weeks 10/19/14 CMIO Proposals Vendors Within 2 weeks of RFP 11/2/14 CMIO Decision Selection Group/CMIO Within 2 weeks 11/16/14 CMIO Contracting CMIO 1 month 12/16/14 CMIO Product Vendor 1 month 1/16/15 Director IT Interfacing Hardware/ IT 1 month 1/16/15 Director IT Software install Training Vendor/IT 2 months 3/16/15 Director IT Go Live ALL Conclusion of training 3/16/15 CMIO Evaluation/ CMIO Ongoing 3/16/15 CMIO Holdback Pmt
6. Commitments Commitment Responsible Target Start Target End Date Date Coding Champion Director Business 1/16/15 ongoing Coders Coding Champion 1/16/15 ongoing IT technical Staff Director IT 12/16/14 1/16/14 Vendor Support/train Vendor 1/16/15 3/16/15 Vendor Evaluation/retrain Vendor 3/16/15 4/16/16 Final acceptance CMIO 4/16/16 4/16/16 9. Roles and responsibilities The group will collaboratively produce a request for information specifying needs regarding NLP and CAC. This will include specifications and contacts with current EMR vendor. Based on information received fro RFI an RFP will be constructed outlining The project requirements, timelines, expectations and acceptance criteria. The group will then negotiate contract, including price, acceptance criteria, training, functionality and support, upgrade and ongoing licensing fees. Finally, the group will oversee install, training and implementation of the product along with ongoing monitoring to ensure acceptance criteria are met on-budget, on-time; meeting the expected objective of > 20 improvement in coder productivity. In the case that the threshold is not met will work with vendor regarding contractual remedies to the resulting deficiency. It is not enough that the product be installed; it must meet
acceptance criteria. Final payments will be made upon meeting acceptance criteria (holdback in contract). Ongoing monitoring to ensure continued improvements in efficiency and revenue.
References Autocoding and Natural Language Processing, whitepaper - 3M, http://multimedia.3m.com/mws/mediawebserver?mwsid=sssssuh8gc7nzxtuo8 _B58mGevUqe17zHvTSevTSeSSSSSS--&fn=3M_NLP_white_paper.pdf. Not all NLP is created equal, whitepaper Optum, http://www.hfma.org/brg/pdf/wp- Not%20all%20NLP%20is%20Created%20Equal_FINAL_06-13-2013.pdf. Nuance Clintegrity 360 Computer-Assisted Coding Gains Rapid Market Adoption with more than 40 New Customer Wins, Nuance, http://www.nuance.com/company/news-room/press-releases/nc_029677. Dimick, Chris. ICD-10 Part of Achieving Healthcare s Triple Aim. Journal of AHIMA website. April 22, 2013. http://journal.ahima.org/2013/04/22/icd-10-partof-achieving-healthcares-triple-aim/. Cassidy, Bonnie. Ten More Questions for CAC Vendors. Journal of AHIMA website. March 1, 2013. http://journal.ahima.org/2013/03/01/ten-more-questions-for-cacvendors/.