Disease Burden Study

A recent question came up:  "What is the best way to conduct a chronic disease burden study?"

Interesting question.
You want to measure the burden of chronic disease.

First, the data.  If you can find good quality data that already exists, this will greatly simplify things... however, in most cases, you will not have access to good quality data.
Good data will be comprehensive (cover the entire population or a representative sample), reasonably current, accurate, and have sufficient detail.  It will need to be collected at the individual person level.

Some possible sources of data:
Surveys:
- If there is a recent DHS (Demographic and Health Survey) this might have enough detail on chronic diseases.  Most DHS data is of good quality.  However, you might not have access to a recent DHS since they are usually only done at intervals of 5 to 10 years.
- Other organizations may have done health surveys and may give you access to their data.  You will have to canvass government, NGOs, and aid agencies.

Routine data:
Some routinely collected data such as that from health facilities can be useful if it has detail on chronic diseases.  However, there are several problems with routine data.  It may not cover the entire population.  There is usually no way to track individual patients to come up with a person count as opposed to a visit count.  There may be other distortions in the data due to poor recording or patient selection.  If you can find routine data that covers the population comprehensively, there are ways to adjust for multiple visits.  However, you still may miss people who don't visit the facilities.  If there are community outreach workers, they may have data which will give better coverage.

Collect your own data:
You will probably be faced with the task of collecting your own data if you can't find existing good quality data.  You will need to do a survey and it does not need to be a large survey if you choose your sample carefully.  A small representative sample can give good quality results.  The sample should be chosen to be representative of all components of the population, age, sex, location (urban, rural), socioeconomic, etc.  A good place to start is the national statistics or census office who often can help you with designing a representative sample.  You should design a questionnaire which is as short as possible and tightly focused on the diseases of interest.  Consider using biomarkers (measure such things as blood pressure and blood tests) if you have the budget.  Here again, community health workers may be a good resource to assist with the survey. 

I assume that you will be using one of the common methods to calculate disease burden such as DALY (Disability Adjusted Life Years).  DALY results in a number which measures the impact of disease by taking into the degree of disability for a disease as well as mortality.  It is also discounted for future years of loss which is appropriate since the society has the opportunity to compensate for this loss.  The DALY is the most useful number for planners to use when deciding on allocating resources if they take into account the cost to prevent and cost to treat each disease.

Hospital Coding - The ICD Conundrum

We Need Good Hospital Data

Developing countries need timely, accurate data on patients admitted to the hospital so that they can understand the hospital demand and patient problems and use this data for planning and management as well as improving clinical care.
How can we get good hospital data?

The ICD Problem

The WHO ICD coding system is a magnificently comprehensive nomenclature for precisely specifying any disease or condition for any patient.  The ICD-9 contains about 15,000 codes and ICD-10 expands that to about 90,000 codes.  It is used extensively in the US and other developed countries to code hospital and clinic visits for reimbursement, patient care and research.
WHO would also like it to be used in developing countries and has been sponsoring training sessions for years.  These sessions typically take a few weeks and give coders a basic understanding of the system. 
I have seen the data produced by these coders in many developing countries and I have never found any data that could be used for planning or management.  In some cases the quality of the coding is reasonably good but it covers on a very small (unrepresentative) set of patients and is months to years late.  In other cases it is just very poor quality (and also usually very late).

A Proposed Solution

We are considering this problem now in the country where I am currently working.  They currently have no data from hospitals.  WHO has trained about 20 coders but none of them are doing any coding since they don't feel competent to do ICD coding and they have been assigned other tasks. There are 19 hospitals but the coders (who have not drifted away) are all concentrated at the national referral hospital.
We need a coding system which can be used effectively with minimal training and which gives timely and reliable information about patients in the hospitals.
I took a look at the DRG coding system and decided that this could be used as a starting point for a simplified hospital coding system.  The DRG codes were designed for hospital reimbursement and they lump a large number of diagnoses codes into less than 1000 different codes which are assigned a "length of stay (LOS)" and reimbursement factor.
I went through these codes and simplified them further.  For the most part, I eliminated the codes which accounted for severity and complications subsets  of the basic disease category.
For instance, most of the codes appear as a set "with co-morbidity and complications" and without.  These distinctions are important for reimbursement but not for basic hospital data.  To keep the coding simple within the expected capabilities and training available in developing countries, these were consolidated.  They can be added back as coding skill improves.
I ended up with a list of less than 300 codes which are comprehensive of the expected hospital admissions in developing countries.
It is also much easier to code using this limited set.

Example

Here is the simplified set of codes for the Respiratory System:
Surg    Major Chest Procedures  
Surg    Other Resp System O.R. Procedures  
   
Med    Pulmonary Embolism  
Med    Respiratory Infections & Inflammations  
Med    Respiratory Neoplasms  
Med    Major Chest Trauma  
Med    Pleural Effusion  
Med    Pulmonary Edema & Respiratory Failure
Med    Chronic Obstructive Pulmonary Disease  
Med    Simple Pneumonia & Pleurisy  
Med    Interstitial Lung Disease  
Med    Pneumothorax  
Med    Bronchitis & Asthma  
Med    Respiratory Signs & Symptoms
Med    Other Respiratory System Diagnoses  
Med    Tuberculosis

The steps in coding a hospital admission would be simplified to:
1. Determine the organ system (i.e. cardiovascular, pulmonary, GI tract, etc.) (There are 25 of these in the code set.)
2. Determine if it was a medical or surgical admission (Was a procedure performed?)
3. Choose from the options within that group (there are between 3 and 15 options at this point).

This would greatly simplify the coding.  It would also give very useful information about hospital use and patient conditions.  Over time as the coders become more experienced and trained, the ICD codes could be introduced for the more capable coders.

For example, if a patient was admitted with "pneumonia" the coding process would be:
1. Determine organ system: Respiratory System (Set 4)
2. Medical or Surgical: Medical (no procedure performed)
3. Then choose from the (14) codes in this set: "Respiratory Infections and Inflammations" (code 179)

This is much easier that performing a full ICD code determination and it should be possible to train the coders to do this much more quickly than ICD coding.  The experience with this coding will also help start them on the path to ICD coding which will require much more extensive training.  I think that this DRG derived coding is a good first step (crawling) before ICD coding (walking).

The Codes

If your are interested in the details. I have uploaded my spreadsheet which contains the full DRG code set as well as the simplified codes.

I'd appreciate comments from anyone.

 

DRG Derived Codes