A guide to successful CDI software initiatives for Ontario hospitals

Over the past decade, Clinical Documentation Improvement (CDI) has become increasingly important as Ontario has moved into data-driven funding mechanisms. With the new focus on financially incentivizing performance across the health system, data quality will continue to be a critical factor as it has been in similar models in other countries.

We’ve had the opportunity to work with over 40 hospitals in Ontario to help improve their clinical documentation through the implementation of our Data Quality Assist (DQA) CDI platform. We’ve published this with the hope that it provides insight into common areas that should be considered by Ontario hospitals to help ensure CDI success.

There are six high-level areas of focus that we will cover:

DATA COMPILATION

The operational challenges of CDI can be broken down into two main categories:

1. Comprehensive collection and compilation of data
2. Accurate interpretation of data

Without a data compilation strategy, it is simply not possible for clinical coders and automated systems to be fully effective.

Any robust CDI solution needs to include a frictionless way of introducing new data into the process without requiring substantial technical integration effort or customization from vendors. We have found that a wide mix of data sources, from automated HIS data feeds and CDI data capture applications to uploaded lists of manually compiled data, have their place in this process.

One thing became clear during the design of DQA: hospitals need to be able to easily bring in their own data without external vendor assistance. The inclusion of well-designed data interfaces is crucial, otherwise a CDI solution will stagnate.

Coders often perform their jobs by accessing a wide variety of screens across many different HIS systems when coding an encounter. Sometimes it’s not even possible to gather all necessary information from relevant data sources and case review becomes a labor-intensive investigative exercise.

To address this issue, we’ve introduced the DataHub module to assist with initial coding. Since compilation of data has always been a key part of the DQA data quality audit process, we now make this data available to the coders at the time of coding to improve baseline data quality. DataHub will continue to evolve to also assist with the accurate initial interpretation of data.

ACCURATE INTERPRETATION OF DATA

Ideally, all data pertaining to a patient encounter would be compiled, interpreted, and pre-coded automatically. At the very least, all information would be visually presented to a coder at the time of coding in order to give them a complete understanding of the care provided. These important sub-groups of encounter data need to be considered:

1. Interventions (CCI)
   CMI and volume funding criteria are based on properly documented interventions. While larger procedures that drive case mix grouping are unlikely to be overlooked, other interventions such as vascular access devices and feeding tubes can be missed. We have had considerable success using DQA cross-reference rules against various hospital data sources to capture those missing interventions when they cannot be discerned from text data supplied at the time of coding.
2. Diagnoses (ICD-10)
   Proper diagnosis attribution is important as it affects CMI and other key calculations. There are many systems that employ a variety of Natural Language Parsing (NLP) techniques to parse data and recommend (or even pre-code) ICD-10 codes, which reduces initial coding effort. But like all machine learning algorithms, there will always be an error rate and they are only as good as the data they are fed (see the “data compilation” section above).
3. Key methodology factors
   Certain data points (e.g. post-discharge home care, birth weight, patient services) have specific implications in Ontario. A physical stay in the ICU affects case weight and is, therefore, a critical factor in documentation, but is not driven from ICD-10 or CCI codes. Similar to interventions, cross-reference data quality checks within DQA have yielded good results.

REGIONAL STANDARDS AND POLICIES

While identifying the existence of the factors listed above is important, it is also critical to account for factors pertaining to Canadian coding standards and provincial policies that affect hospital funding and accountability agreements.

As an example, the case weight attributed to an encounter is highly dependent on choosing the proper most responsible diagnosis (MRDx) which can sometimes be difficult to determine for complex cases where the admitting diagnoses ultimately doesn’t represent the underlying care and resource utilization associated with the encounter. Other types of minor coding mistakes can directly affect the eligibility for volume funded cases, directly affecting hospital revenue.

These Ontario-specific rules and calculations must be deeply integrated within the CDI tools to ensure that proper attention is paid to the factors that have substantive impact on the hospital. Coding teams have limited resources and they need to focus on what matters most. This was very apparent during the initial rollouts of DQA when we observed that almost every hospital had built their own rules and reports to directly address these provincial policy factors, and that the maintenance effort involved had often become cost prohibitive.

AUTOMATED CASE REVIEW

Despite best efforts, mistakes can be made during the documentation and coding process. Sometimes a physician doesn’t properly record a key factor, something is misinterpreted, or key information isn’t available at the time of coding.

Automated case auditing has been a staple of the DQA CDI platform since its first version. With the help of many hospitals over several years, we’ve designed case profiling logic that uses a variety of statistics and Ontario/Canadian methodology calculations to determine whether a coded abstract is abnormal. We combine all factors to produce an aggregate outlier score that has yielded very positive results for documentation problems that are difficult to detect.

It is also possible that a hospital chronically under-documents patient care in certain areas. It is particularly difficult to identify a widespread absence of documented information. Our approach has been to analyze peer data (e.g. provincial case costing data) to generate statistics on occurrence rates of CMI factors for similar cases at peer hospitals. If there is large discrepancy in these occurrence rates, it generally points to variation in care practices or undercoding on those factors.

PROCESS EFFICIENCY AND ANALYTICS

Supplementary functionality to support the case review process is critical to minimize the input costs of CDI. Examples within DQA include:

1. Root cause analysis through the use of analytical reporting to identify repeated systematic issues that can be remedied by changes in process or physician education.
2. Collaboration tools that support the case review process including centralized case notes and workflow assignment to team members involved in the review process.
3. Effectiveness monitoring and rule tuning. In addition to enabling rule tuning to reduce false positives in automated case reviews, detailed statistics on case corrections allow hospitals to directly calculate the impact of their CDI investment. Without this level of detailed auditing in the CDI system, any attribution of benefit to the system can be confounded by unrelated, coincidental factors.

Beginning in 2019, we introduced a new level of analytics for our clients to help explain year-over-year variations in inpatient weighted volume and CMI for volume-funded cases. Finance teams need to understand the underlying root causes for CMI change beyond data quality and this can include many factors including occurrence rates of long stay discharges, variations in conservable days, and changes to patient case mix.

EXTENSIBLE DATA QUALITY RULES

Flexibility in creating new data quality rules has been a key driver of DQA’s success. We often say DQA was built by coders across the province because, in many ways, it was. The flexible rule engine allowed us to integrate new intelligence rapidly based on techniques and approaches proven by Health Records departments across many hospitals. This also ensured that rules specific to Canadian and Ontario coding were properly represented in the platform.

In addition to incorporating novel data quality rules, this flexibility is absolutely required to adapt to changing governmental methodologies. This can be a problem with generic CDI solutions adapted to the Canadian environment, because it is difficult for international vendors to evolve their products at a pace that matches change in local policy.

HINTS AND PITFALLS TO AVOID

1. Be targeted in your efforts. Hospitals have limited resources and need to focus on what matters most. Targeted CDI initiatives that address critical factors will always provide more benefit than generic approaches to data quality where real value is a coincidental by-product.
2. Approach adapted solutions from the U.S. or other countries with caution. They can make great tactical components but won’t address the entire strategy. International vendors have a massive hurdle to overcome: they often don’t adapt their software to address regional issues in smaller markets and instead focus on features that are easily replicated between countries. This means they may not identify data quality issues that are the most important to your hospital, which is very dependent on government policy.
3. Similarly, do background checks on vendor effectiveness before accepting astounding claims of massive CMI jumps, particularly if they haven’t comprehensively analyzed your baseline data quality. Investigate the veracity of these assertions by speaking with those peer hospital clients in the province. CMI increases are easily verifiable so make sure things add up. Also, results from other countries are unlikely to be appropriate comparisons since CDI challenges and benefits differ significantly across countries.
4. Using machine learning algorithms to automatically identify factors such as ICD-10 codes can certainly streamline coding but it comes at a cost. Overreliance on automated systems diminish scrutiny by qualified coders and can introduce additional errors and omissions. Finding the right balance is key.
5. Be wary of claims that retrospective review can be made obsolete. Great processes and coding assistance tools like DataHub can help get you to 99% CMI accuracy, but automated coding audits with full case coverage help get you that valuable last 1%.
6. Don’t ignore integration challenges. Robust CDI processes need as much data as possible. CDI solution providers must be very familiar with typical Canadian data sources within hospitals to ensure proper integration.
7. If you’re one of the many hospitals undergoing an EHR migration, CDI is even more important. Existing systems and data feeds to clinical abstracting systems will be modified or completely replaced and that introduces new risks to baseline data quality.
8. Beware of long, expensive implementation processes. A good CDI vendor should be able to demonstrate value without risking large sunk costs and no way to turn back if expectations are unmet. Helping hospitals manage their risks by getting them up and running within a week and delivering immediate value has been a large factor in the rapid adoption of DQA.

SUMMARY

When planning to implement new software as part of a CDI initiative, we recommend using this article as a guide to ask specific questions of your team and vendors to determine whether proposed approaches address a few of these issues or the problem in its entirety.

Comprehensive CDI is nuanced and there are no silver bullets. Without taking a holistic view of the problem, hospitals run the risk of undertaking a large project for something that solves only one small portion of the actual problem.

For more information about implementing the DQA CDI platform, please contact us.