January 25, 2011 CBCC Conference Call
Community-Based Collaborative Care Working Group Meeting
- Bill Braithwaite, MD
- Ed Coyne
- Floyd Eisenberg
- Suzanne Gonzales-Webb CBCC Co-chair
- Mary Ann Juurlink
- John Moehrke Security Co-chair
- Ken Salyards
- Richard Thoreson CBCC Co-chair
- Ioana Singureanu
- Serafina Versaggi
- Tony Weida
- Craig Winter
- (05 min) Roll Call, Approve Minutes (from January 25 call) & Accept Agenda
- (45 min) NQF measures review with Ioana / Floyd Eisenberg
Action 1: Prepare questions regarding NQF measures for next week’s meeting
- Update: Questions sent to Floyd and coChairs on Friday Jan 28, 2011
Guest: Floyd Eisenberg. A Physician, current Senior VP Health Information Technology at National Quality Forum(NQF), previously at Siemens as an Inpatient Software Developer, prior Chair for Population Perspective Technical Committee in HITSP and a Chair in IHE Quality Research and Public Health Domain.
Ioana: We’ve analyzed the behavioral health specific NQF measures for 112, 103, 105 using the Quality Data Set (QDS) and we’ve tried to understand what information would need to be collected for the EHR system so they could compute these measures, ideally in an automated fashion directly from their clinical data. We would like this data to be more outcome based, right now it is more process based. With the process specific measures we would like to have it as much as possible to resemble real life clinical data. We are hoping that Floyd will guide us to understand the relationship between this data and which relationships are important to compute the NQF measures as well as what we should be looking for in the future.
Floyd: The NQF sponsored the Health Quality Measure Format (HQMF) as a draft standard in HL7. Also through Health and Human Services mostly through CMS we retooled 113 existing measures into an electronic format to address data within the EHRs. We would like outcome measures but one of the challenges we had is that we were restricted to deal with the measures as they were endorsed and as they existed. All these measures are process measures, no outcome measures. Because of that we looked for data in the EHR that is real clinical data avoiding anything that was a G codes (G codes are used to report quality measures under the Physician Quality Reporting System), to get around clinical data to submit as a claim or a CPT 2 code which is basically attestation. It was trying to retool what existed as a process measure for each of these to find data and represent it in a way so it could be used in an EHR. Now that is in understanding that physician and other clinician workflows didn’t necessary enter data in this way, and every EHR didn’t necessarily capture it. What we tried to address was where we would find the clinical data. Going back to what occurred in HITSP in determining C32 and C83 and 154 (HITSP constructs) as well as what the standards committee recommended and that came out in certification rules. Every EHR requires problem list, the intent was to look for data in a problem list not a claim for example we looked for those meds that were on an active med list either ordered or dispensed. What we used was a model of information that we called Quality Data Set, which was renamed the Quality Data Model. It is not so much a list of elements but a model of how the elements are represented. The model basically uses the concepts e.g. medication and its context of how it is used e.g. administered, dispensed, and active or allergy to the medication (reason to exclude a patient). In order to address the vocabulary used we tried to deal with what was required for certification for EHRs and what was suggested for future e.g. ICD9 for conditions, 1CD10 which is a 2013 requirement for claims, and SNOMED when looking forward to where things will be. That is why you will see different taxonomies together, rather than putting them in one value set / encode list we made it a grouping of 3 different value sets / encode lists. In order to identify where that information should be we relied heavily on CDA.
Ioana: We are reviewing the full XML versions of the retooled NQF measures sent to us by Ken.
Ioana: We are creating a quality data model, in other words a subset of the EHR data that you will need to compute the NQF 0004. We want to be sure that the EHRs are capturing this data. We are looking at the properties of these objects that you have identified in the NQF measures e.g. NQF 0004 and are using it to compute the numerators and denominators. A patient person would have more attributes in the real world but the only thing we might be interested in for a particular measure is the age so we would only have a DOB to determine, for example that they are adults. We are only looking at the requirements that we need to compute the measure nothing else. We are focusing on the subset identified in the retooled measures.
Floyd: One challenge we had in doing the retooling was in understanding how can we describe the relationships between elements and when and in what sequence they occurred. Bob Dolin provided a set of comparators that he had balloted through HL7 and advised us that this was really all that there was to use and that is what we used in the XML portions.
Ioana: What we are struggling to understand is, identifying which subset of data would qualify a patient to be part of a specific population. For example if the diagnosis was identified during the visit or is it already part of the problem list. Similarly are we talking about a procedure that occurred during the measure period or are we looking at procedures that occurred during a visit.
Floyd: The NQF 0004 Initiation and Engagement of Alcohol and Other Drug Dependence Treatment is looking to identified someone who was recently diagnosed, >45 days before the end of the year given the physician time to do the intervention.
Ioana: So the measure is looking for the date this diagnosis was assigned by a clinician and must occur 45 days before the measure period ends e.g. effective date. A point of validation in order for us to captured diagnosis is to have:
- Diagnosis code associated with substance abuse per the measure e.g. value set associated with the diagnosis code
- standards code e.g. we are looking for active diagnosis
- effective time it occurred e.g. between a specific time period
Floyd: When we are thinking about a problem list this assumes reconciliation of a problem and use of a problem list the way it is not being used today. With respect to NQF 0004 it is assuming that when the problem was initially active on a problem list this is the concept of activation effective time, and assuming that it had to happen sometime during the year and it also had to happen during an encounter during that year.
Ioana: This is where the association of problem list comes in
Floyd: If the doctor is seeing the patient and not billing the encounter, do we really need the encounter or can we say a patient of yours has an active diagnosis, but they use an encounter to determine that the patient is that doctor’s patient. They use it not only to say it occurred during an encounter but to make sure they attribute the right diagnosis to the right doctor.
Ioana: One way to read it is a person had a visit during the measure period and during that visit things happened. If those things that happened meet the criteria then this person is added to the population. Another way to read it; there is a person and during that year has had a new diagnosis assigned to him, and whether it happened in a visit or not we don’t care but for the purpose of this quality measure we need to know the effective time occurred 45 days before the end of the measure period and one day after they started the measure period. That is what makes this particular person eligible to be part of the population criteria we are computing. Alternative path and which one is the right one?
Floyd: One of our challenges here is based on claims and was dependent on encounter rather than ‘during the year’. It would be a substantive change to say as long there is one encounter during the year and the diagnosis occurred during the year than that person is eligible. They had to tie it to an encounter. I like the alternative model better I don’t know if that is easier for the EHR or not.
Ioana: In terms of building the population set and counting the population, I see the appeal of saying a patient has had one encounter during the measure period so we include them. Then let’s look at those who have not had an identified diagnostic and let’s exclude them. This is the way the numerator and denominator logic build. This is the way we are building the model, you’ve not has a diagnosis or if you had a diagnosis did you have a procedure e.g. detox, counseling during the measure period. It assumes everything had to stem from a visit from the provider.
Mary Ann begin_of_the_skype_highlighting end_of_the_skype_highlighting: This is true for this particular NQF but not necessarily for each NQF. We need to have a domain model and constrain for each applicable NQF. We need to ensure we have the right buckets so we can capture the right information.
Ioana: We are analyzing each measure independently, but with all the knowledge we are deriving from this we will be putting it into a single model that would be able to say ‘here is the EHR model required to compute the Behavioral Health measures. If we can answer these questions we can build the relationship as close as possible to the intent of the measures. We want to be able to build this model to align with the intent of the measure. If it turns out that implementers can’t implement because they are not tracking everything by encounter then this could become a point of discussion as to why people are not able to use the measures. If we want to have people use the measure correctly we want to put forth what we expect to be the correct relationship. What we are really trying to get at for this measure is whether the encounter occurred or not, we want to know if the diagnosis was put on the problem list during the measure period and we don’t care how you get there
Floyd: If this type of view is important to implementers how should it be included in the eMeasure at HQMF as opposed to the way it is appearing? The concepts you are coming up with implementation are going to apply to more than Behavioral Health measures and there is talk about doing a domain analysis model for HL7 May Ballot. So all that you are doing with the SHIPPS project will need to be incorporated as well.
Floyd: There is significant interest in creating an implementation guide around HQMF which may be a domain analysis model in HL7 (Bob Dolin, Patty Grimes, Freida Hall, Stan from CDS and Structured Documents).
Ioana: I see a difference between the two models (SHIPPS and HQMF) because HQMF represents criteria, so whatever information you need e.g. a visit or diagnosis HQMF is agnostic. It says plug in your discriminators and we will compute the population. So the two models are complementary. The information SHIPPS is encoding is a way to instantiate the HQMF. In order to instantiate a behavioral model you need to understand what information you are dealing with before you instantiate it correctly. The HQMF will validate, it will validate even if we put the wrong information in. SHIPPS would be an add-on to what the other group is doing with the HQMF. I assume they will be approaching it by domain.
Mary Ann: Question: what information should we be using e.g. retooled or original work  ? For example the information NQF 0004 negative 60 day period for diagnosis history which will determine a new episode is missing in the retooled version.
Floyd: The negative 60 day period for new episode was dropped inadvertently. It is only considered first if it was not active for 60 days. It should be an exclusion and not in the denominator.
Ioana: As part of QA after we complete the model we will restate the criteria in the measure in these terms so you will have patient person, diagnosis etc. At the end we will have the representation of the criteria in terms of the QDS concepts. This will validate that we are on the right track.
Floyd: These are going out for comment and you have just identified issues that will have to be fixed when it goes to final publication
- >= 60 days needs to proceed ‘the before simultaneous to’ and it has it in the denominator as an ‘and not’ but the 60 day on the comparison is missing.
Mary Ann: The process looks something like this;
- Understand the intent of the measure
- Review NQF home page
- Review retooled NQF measures
- Outline salient points
- Understand the information needed to compute the measure
- Understand the relationships
- Review any additional information in the retooled denominator / denominator and in the criteria for data being captured
- This information we are not capturing in the model
- We will capture it in the structure queried statements
Ioana: We need to review the numerator / denominator and criteria in order to provide meaningfull use of the measures.
Floyd: If you do see a discrepancy in what we have delivered and the original intent if you can list those out that would be extremely beneficially for us. Your careful review will certainly help. The comment period is 60 days Feb 01 and Apr 02. All 113 will be out at this time. The PHQ 9 will also be out.
Mary Ann: the PHQ 9 is an assessment scale and how will we be using it. In a lot of the NQF measures there are various assessments so what are we interested in? Are we interested in the score, the detail, trending etc? Should we be looking at creating an assessment scale model that will incorporate all assessment scales?
Floyd: The PHQ9 is looking at a series of 3 things.
- Was the assessment completed?
- After stating the data element, we state in brackets what we looking for, is the attribute of the result value
We define it as three values
- Initial measure e.g. value = 9
- For all the results we are looking to see if there was a second measure within 6 months value =5
- Follow up value instance a 11-13 months value = 5
We handle the result as an attribute of a test or a procedure, intervention. We listed the attribute in a set of brackets after the name of the term to try to keep to convention. If the result has to be specific value we would use SNOMED or in the case of xRay we’re looking for results consistent with pneumonia, there could be a whole list of values and if any present you include if not you don’t include. So what we did was to use the brackets after the data element, to indicate what we are looking for. So the result value here is numerical and if there is nothing written it is not present
Example residence attribute (nursing home) should this be location vs a characterization of where they are. What’s the better way of what we are trying to do this? This may be a comment on the quality data model itself.
Serafina: Where it says diagnosis active major depression and the ordinality of principle I am assuming it occurred in this particular encounter and that it was the primary diagnosis. Does ordinality mean the sequence of all diagnoses that may be associated with this encounter?
Floyd: Part of ordinality it is in-patient billing anomaly.
Serafina: Why I put priority code in diagnosis, which might not have been relevant to NQF 0004, however both the diagnosis and the procedure for inpatient episodes where you are going to have that called out in a measure we need to know the priority code in the RIM, what ordinality it is within the diagnosis and or procedure coding system for inpatient type anything that is done by an inpatient like ambulatory visit.
Floyd: This was a general measure issue, by using measures ‘as is’, we couldn’t get measure developers to agree that if it is significant and on the problem list does it matter if it is principle or not shouldn’t you be doing the same thing for that condition because the patient has it. They were very adamant of staying with principle.
Serafina: It is all based on how the reimbursement works. In how providers are doing care and therefore how systems are documenting care.
Floyd: Unless we can move documentation we are stuck with the billing paradigm that is why you will see principle in many of these measures. What we really need is a way to say active condition and of that activity what is the significance of it is related to why I am seeing the person or it’s a high likelihood or low likelihood and there is no way of doing that right now and represent it, which is why we are staying with principle.
Richard: You said earlier you probably had to do some things you didn’t particular want to do because you were doing the retooling from billing.
Floyd: Looking at the inpatient ones what they want to know is there a presumptive diagnosis of pneumonia in the EE then once they get to the floor that’s where you look to see they have the right antibiotic. The question was how do you know if it on a differential diagnosis, how high on the differential represents presumptive and I don’t think there is a good way to represent that right now.
Richard: The quality measure for whom. If you are measuring me a one physician is one thing if you are measuring me as a hospital is another if you are measuring me as a whole network of people in a health care home is another.
Floyd: You are right how we are dealing with attribution makes a difference too. This is part of the challenge as these are intended for meaningful use so they have to be attributed to a provider in this case eligible provider measures. These are based on the way the measures work, without going back to the steering committees that developed the measures some things they could not change.
Measure developers: AMA & NCQA
Latest definition list published in September 2010 Quality Data Set. The next publication is March 2011.
Meeting was adjourned at 2:58 PM Eastern