Navigate: Home| Mission Statement | Background | Staff and Training | Glaucoma | Diabetes | AMD | Contact Us | Related Links



Interested in Grading

In this section of the website we aim to provide you with a strong overview of the grading process. Please read on, or alternatively click on one of the links below to go directly to your chosen area.

Reading Centre | Grader Training | Logistics | Quality Control | Protocol and Time | Problems with Grading | Summary | Data Management Information

Reading Centre
A Reading Centre is staffed in the following areas, a lead clinician, principal investigators, senior graders, graders, graders in training, epidemiologists, statisticians, photographers and other staff as needed. The lead clinician and principal investigators are ophthalmologists.

Grader Training
Graders have no medical background, but are usually educated to degree level. A senior grader trains them until they perform to a certain standard. Graders must sit a grading exam at the end of their training. Overall training is for a period of between four and six months with additional specialised training sessions. For a grader to receive certification they need to grade a certain set of exam photographs, where grading has been agreed by the senior grader and the reading centre advisor. They need to grade this set at least twice with a 14-day gap in between. The grading is then entered onto a database and intra and inter-observer agreement are calculated. If this is acceptable a grader can grade alone with some supervision. If not acceptable, further training is followed by re-examination. Graders are generally certified to grade for a specific disease and not as a grader. Currently the problem with grader training is that no universally accepted training schedule exists. In addition there is the question of; who certifies graders in areas when there is no accepted grading form/severity scale.

Logistics
After the protocol has been agreed photographs are taken and sent to The Reading Centre. On their arrival the images are checked and catalogued. They are then kept under lock and key for the entire time that they are held at the Reading Centre. Graders do not have access to the photographs and are given the images on a daily basis by the data handler. Once graded, images are given directly back to the data handler and then once again locked away along with the completed graded forms. A
s images are always double-graded this process ensures that no grader has access to previous grader results. The data collected is then entered into a secure database which is stored on a separate drive, no-one other than the data co-ordinator and IT staff have access to this drive.

Quality Control
For quality control purposes a subset of photographs from each study is double/triple graded and entered in to the database. Again, 14 days has to elapse before a grader is allowed to regrade images they have previously graded. This ensures that graders will be unable to remember or memorise any of their prior grading results. Subsets of photographs are graded by the same person, entered, analysed and discussed. A separate set of photographs are kept for temporal drift, they are graded every year, entered and never discussed in detail. If results of this set are not up to standard then further tuition is given.

A major problem with quality control is that all reading centres currently do their own. This makes it very difficult to assess studies. There is a push for specific quality control procedures or image 'swapping'.

Intra and inter observer agreement can be described in three ways; exact agreement, kappa values and weighted kappas. When presenting these to the graders, specific explanation is given about how they influence the overall results.

Protocol and Time
When a protocol is developed for grading it needs to be categorically decided and stated what we are to be looking for when we grade. Points to consider are; need to be able to answer the proposed question, needs to be specific enough and needs to be quantifiable and reproducible. Most importantly, a study's needs have to be matched with the centre's capabilities and needs to be completed within the given time-frame. A rough guide to how many images can be graded within a specific time are given below:

Experienced Grader; 10 AMD photos/one hour, 3 Diabetic Retinopathy/one hour, 3 FFA/one hour.

Not Experienced Grader; 3 AMD photos/one hour, 1-2 Diabetic Retinopathy/one hour, 1 FFA/one hour.

For concentration purposes and for maximum accuracy a grader should only be allowed to grade for between 4-6 hours a day.

Problems with Grading for Different Studies
In general there are still problems that continue to occur with grading for different studies. As yet, there is no universally accepted grading form and no universally accepted fundus photography system in existence. It is also a concern that there is no official programme for grader training and quality control. When grading for screening purposes, such as diabetic retinopathy, there is no information if high-risk or end stage disease patients are seen or not. These all need to be kept in mind when analysing grading results.

Summary
To summarise the process of grading an important point is to aim before you grade. Make sure that you have taken all aspects of the study into consideration and clarified anything that is not clear; it is imperative that a clear and concise protocol has been developed and agreed upon. Make sure that you have your logistics right. Detailed and structured grading is essential. Finally, be certain that you have your quality assurance correct.

Data Management

Why have a database?
All organisations rely on the importance of good data and its availability. A database is basically a filing cabinet for that information which can be accessed via networked computers, intranets and the intranet. It means that information is readily available at the touch of a key. It also means that you are able to be more selective about the information that you retrieve.

Managing a Clean Database: Rubbish in, Rubbish Out
A lot of us are familiar with the saying or expression; 'if you put rubbish in you will get rubbish out.' If rubbish is entered into the database, this will obviously have a great effect on what comes out. Rubbish, obviously being inaccurately entered information, will undoubtedly have a huge effect on the final results of any study taking place. If data is entered badly, the results of the study will be inconclusive. If this happens and data is not double or even triple entered or checked, the likelihood is that the whole procedure will have been a complete waste of time, effort and money. Making data entry as simple and as straight forward as possible is the easiest way to prevent bad data being entered.

Designing a Clean Database - How Easy is it to Enter Data?

Drop Down Boxes
A drop down box works, by giving the data entry person the choice of codes, numbers or words relevant to a particular field. They are useful to doubly confirm a field is entered in the correct manner. Although it is difficult to enter data in this way, as it is very time consuming. It does restrict the opportunity for error, by limiting the available keying options to those stored in the list.

Locking the Database
Locking the database ensures that each field is completed in the correct manner. Although this does not - in any way - guarantee that fields have the correct information entered.

Matching Tab Order
Tab order in the database should match the running order of the data sheet, from which data is being entered. This assists the person(s) entering the data to obtain some sort of running order in which data is keyed. Data entry then becomes consistent, which minimises the risk of mistakes.

Do not 'crowd' the page
If the database has been designed well then a page should not be crowded. By keeping a database sheet clean and simple, it makes it easier for the data entry person to scan for errors after each sheet has been completed.

One database screen, per datasheet.
Data is much easier to enter if the database screen is kept to one sheet only. Anything more than this greatly improves the opportunity for errors.

Automatic Wording or Coding
An example of this is, if a word or code needs to be repeatedly entered, programme likely words into the database so that it completes the word after the first letter is keyed. For instance, if on every database sheet a field needed to be completed with a person's name, the database could then automatically make the appropriate selection after the first one or two letters is keyed.

Summary
We can see now the overall importance of a well-designed database and how careful planning can greatly assist in achieving and providing accurate results. When considering your database design you should also have clearly defined study objectives and outcome measures. You should consider which statistical method you will be using for analysing your data and ideally meet with a statistician beforehand for some practical advice. Be aware of the data availability from your data-collecting source and confirm that all necessary fields will be able to be completed. Entering a small amount of data as a test in the beginning of a study can save a lot of time and heartache in the future. Therefore it is essential to regularly monitor the data and feedback, giving you the opportunity to deal swiftly with any problems or to make any necessary amendments to the database.

The Reading Centre
Moorfields Eye Hospital NHS Foundation Trust