Within practically any industry, key processes hinge upon access to information and data stored within documents and it is no less in healthcare. Even with significant advancements with EHR/EMR and interoperability, use of documents continues to flourish due to a number of factors including that documents are easy to use, and changing from documents to other data formats is difficult to pull-off. Many automation solutions choose to ignore a transition at all, preferring to have clients start from day-one not using documents. The reality is that documents proliferate because of their ease of use and sharing.
Even still, the organizations that make-up any ecosystem do not have to transition from documents in order to gain the benefits of automation. Rather, adoption of intelligent document processing (IDP) software can provide both a short and long-term path to digital transformation and complement existing and future plans for applying new types of automation using machine learning.
One key factor with assessing the opportunity to automate within the healthcare world is to understand the nature of documents you use within key areas. Processes such as claims submission and adjudication, payment reconciliation, and even audits all use similar yet different documents, the attributes of which can have significant impact on the path you choose. Let’s look at the three common types of documents:
The Structured Form
Structured forms are all over healthcare, from benefit enrollment forms to claims. But they are also one of the most misunderstood. Structured forms are defined by the nature of the layout of data (also referred to as “fields”). For instance, a CMS-1500 form always has the patient’s data such as name, address, and birthdate always in the same location and always in the same format. Having such standardization makes it much easier for a person to look at and enter data into a medical claims system; it also makes it much easier to configure an IDP system to know where to look. You might hear the word “template” to describe the layout and the manner in which the software is configured.
With highly standardized forms, the automation potential for structured forms is arguably the highest among the three document types as there is less potential for error with locating information. But there is also the hidden challenge, especially within healthcare: poor quality. While it is entirely possible to have highly standardized forms, it is another thing to standardize scanning quality, especially if documents are shared with fax machines. The expected location of key data suddenly shifts up, down, right, or left. And unwanted noise can make clear-legible data on the original into a smudgy mess. So even while structured forms are the simplest and offer the highest potential for automation, there are still challenges to be solved.
Moving to the next document type, and the next level of difficulty, is the semi-structured document. Let’s use the health remittance (otherwise known as EOB or EOP) as an illustration. This document shares similarities with structured forms: needed data has labels so that staff can find it and perform data entry and if your organization only deals with a few payers, then the data for each payer’s remittance will always be in the same location. What really drives the difference between the two is the potential number of variations of a remittance document. No two payers use the same format which makes it more difficult and time-consuming for data entry staff to locate each required field. Data such as the covered amount might always be there but in a different location for each payer requiring even an experienced data entry clerk to spend time hunting and the more payers a provider organization deals with the higher the variance. The more difficult it is for us humans to locate data means that it is more difficult for IDP software too. Most software uses a technique such as keyword anchors to use as “clues” as to where the data might be. But since there is more variety with both data labels and the location of data relative to the label and where it is on the page, there is a higher potential for error.
The third document type is the most challenging and the reality is that there aren’t many processes that have achieved a high level of automation with these types. Documents such as progress notes, contracts between providers and payers, and lab reports all represent documents that contain a large amount of unstructured data. Unstructured data doesn’t have labels or other clues that enable staff or software to easily find the specific information needed. A set of symptoms or a diagnosis can be buried within paragraphs of text requiring staff significant time to skim each page. To automate the process of locating information within unstructured documents requires yet a different set of techniques using grammatical hints – a range of capabilities often put under the banner of natural language processing or NLP. Even with lower levels of automation, significant efficiencies can be enjoyed just by helping staff to be more efficient at doing their jobs. If the exact data cannot be located, the most likely page where the data resides can be shown or multiple options can be presented within the document leaving the work mostly to approving what the system finds.
The Promise of Machine Learning
Most automation projects using IDP software have historically been targeted at processes that make heavy use of structured forms or a limited number of semi-structured documents. This was largely due to the relative complexity of configuring and optimizing a system to deliver comprehensive amounts of data at a high level of accuracy. Unstructured documents within an automation project were quite rare. But with the introduction of machine learning into the IDP world, the challenges of manually configuring and optimizing a system become significantly easier as the system configures itself using training data sets. And once set in-place, the system continues to evaluate data, continuing to adjust. Even structured forms with high variance due to poor quality scans benefit from automated image enhancement.
The upshot is that getting benefits of automation for all three document types is getting easier but the nature of each document within your project and the relative difficulty will still impact automation potential. You just won’t have to put in the hundreds of hours to benefit.