 So predictors are data element values that are automatically generated from samples of current and past data depending on the periods you have specified to be included in the calculation. So we need to define what is our period that we will select for generating this outlier or a predictor value, what will be the current and the past data for what periods will be used them to calculate the predictors. So for example for reviewing immunization data quality predictors can be used to analyze previous data in order to generate outliers for future data. So this is a little complex and for this academy we will mainly focus on what a predictor is and how it is used but we will not go into the details on how do we create a predictor. But there is this link in the presentation for the manual which will be helpful if you want to further create predictors yourself or learn more about predictors. But I can show you one example on how these predictors work. So say for generating a predictor in DHIS2 we say that we have to use the average for last 11 months and then three times the standard deviation. So suppose this is the way how we are calculating a threshold value. So then the periods would be 11 months that precede the last month. So for example I am generating the threshold for December 2020 then I will skip November 2020 and then last 11 months. So 11 months before November 2020 will be considered. So then we'll use December 2019 to October 2020 in my calculation. So this is the logic that we use for the threshold creation and we'll just exclude any value that is an extreme outlier. So if we look at our predictor. So suppose I have to calculate the predictor the threshold value for December. I do not consider the date of November. I consider the last 11 months which is December 19 to October 2020. The average of these 11 months is 122.8 and standard deviation is 16.4. So then my outlier calculation would be average plus three times the standard deviation which is 173. So this we've already predefined in our application the thresholds for the semenization example that we show. So then in this scenario if my threshold or outlier is 173 that is my predictor's value. Then any value if I enter above 173. So in my December data entry if I enter 180 or 200 then it will show a valuation alert that this is outside the threshold value. Using the generated outliers. Once we have generated the outliers we need to check our data to ensure that our data values do not exceed the outliers that have been generated based on the formula. So we can either use these threshold values which we are generating through predictors or we can import the values from outside say subway data or thresholds generated from some other software. Those values we can import in DHRs too and we can use those values for our validation rules. So the process is same we need to have a data element where these thresholds are stored. These can either be stored by calculating from predictors or these can be imported from external software. And then we perform a manual check by reviewing the actual value versus the generated value. So we'll see them in the application in some time. This is just an example what we'll see. So suppose for Beggar Hospital we see that RV2 doses given, rotavirus doses given to children is 6757 while our threshold that we have set up based on the last 11 months is 700. So then we can say that this value exceeds the threshold value and we need to go back and check this data. So it looks quite obvious and maybe looking at these values maybe there's an additional digit which has been put in by mistake. So this will just help the user to see if there is any mistake, a manual data entry mistake in the data or they can just cross check and correct it. So how do we use these generated outlier values if we go to each months, each periods data entry screen and look for these validation errors? These might be difficult because you have to look at many time periods to assess the threshold values. So this may be difficult however if you want to look at many time periods, organization units and data elements together at the same time for consistency. And it's always easier to run these outliers through validation rule analysis. Though we can, we will see an alert in the data entry as well but since it makes sense to compare these outlier values for a longer period of time, it's more effective if we do this validation check through the validation rule analysis app in data quality. So we stop here to look at how do we run these threshold values in the application while how do we configure these we look at them tomorrow. I think we have a request to describe the predictors again. Next question why no considering normal data for the predictors calculation? So I just understood we have to describe the predictors again and what was the question? Why no considering normal data for the predictors calculation? So I mean that's how these threshold values are configured. These are the standards that are being followed and based on these guidelines, so if you look at the manual, there are these guidelines what periods to skip, what periods to take in the count. So for our example our logic is we for predicting a threshold value, we don't take just the last month's data, we take the 11 months data before that and we predict the values based on those 11 months. That's how the guidelines which is followed for this indicator but I might not be able to drill down further on the conceptual thing of what periods should we consider for what kind of thresholds. I think there are few details in the manual but then we need to refer to guidelines for setting a threshold for the different type of data fields. And just to explain the predictors again. So basically we create a data element and we predict that what should be the threshold value. So say my data element is RV two doses of rotavirus given to children and I have to set up a threshold value just to adjust the quality. Yes please. What if you want to calculate the predictor was a period of math or beyond the event math like we want to create to calculate the predictor for its math well and we succeed by doing so. Yes we can do that so when we look at how we create a predictor tomorrow though I will not go into detail on I mean we'll not practice how our predictors being added but I'll run through on how a predictor is created. So therein we can set up these values that how many periods do you want to consider how many periods do you want to skip in between. So we'll cover so you can define if you want to use just six months you can define that while you set up your predictor. So it means that if you want to calculate a predictor for six months we will not consider the last month maybe if it's from January to June it means that you will not consider June. So if you have to calculate predictor for July so then your last six months will be the months preceding June. Okay. So this same to May yeah okay so just a quick recap on the predictor so this will help us assess what should be the threshold value so we predict these values so we can do a calculation take an average of last eleven months data add on the standard deviation to it and show me what is the threshold so based on that it will say that okay based on your data average and what are the standard deviations maximum should be 700 your doses should not go above 700 otherwise it's an outlier and so you could compare those values with the values. Now these threshold values we have for this example calculated through predictors but if you have these auto generated through some other tools we can import them in the same data elements. So if that's okay to die move to data entry so now we'll see an example of how do we run these predictor generated threshold values and related validation rules. So for this example I have selected Svegel hospital gateway PHC my immunization program and now when I run validation so I see for my DPT dose it shows an error that DPT dose administered is less than or equal to DPT dose threshold. So now if we look at these monthly validations they though they show us that your value is exceeding the threshold values but it's not compared easy to compare them on the data entry app it makes more sense to go to the data quality and run validations there. So if I go to data quality and try to run this validation step go to validation rule analysis and data quality then select animal region for like last three four months and I take the immunization thresholds validation rules. So here then if you've made these rules for all it will also run if there are zero values but let me show you one where we have these thresholds. So if you look at this example here I will look at the details. So rotavirus 2 administered is less than or equal to rotavirus threshold wherein it says that the dose is given is less than I think I've clicked on some other I'll just show you one more example which is more clear than this one. Okay so I think we have data for January here let me just go back two more minutes. So here I think this is not showing value there is some error on the message but otherwise you can see that there's an error in the validation rule my threshold is 700 and my value is seven one three which I've entered if I go into the details it says this is a validation error for rotavirus 2 administered wherein it should be less than the mean and three standard deviations of last 11 months and I could check for this data if this is an outlier or a data entry error and I can correct this data there. Since there are a lot of questions on predictors I will although we discuss this in detail tomorrow but I will just show you how are these set up. These are the predictors which we can define in the application based on these formulas that we have so if I open this predictor for DPT threshold values so I can say that every per every month I am calculating the threshold so the threshold values keep on changing for every month based on last 11 months and here in we define that how many months do we have to calculate so sequential sample count means I have to take last 12 months and I have to skip one month so that's where I'm saying sequential skip count so for this formula I just take the 11 months preceding the last month so take 12 sequences and skip one and then if I go to my generator I have the same formula I'm taking the average for the doses given in the last 11 months preceding skipping the last month and three times the standard deviation from the population where dose is given we'll come back to this tomorrow but then this is where you could set in what are the number of periods that you have to take how many periods you have to skip if you don't want to skip the last month you can just set it to zero there and the steps are same so if you look at your exercises now where the exercise 4 and 5 are for the thresholds so you will be able to see where are the errors in the thresholds and which will be further clarified once you go to the validation rule analysis and generate these so are there any questions if we take an example of where we were ourselves adding the value so even then we'll have these data elements where the thresholds will be stored we can just store those thresholds and once we run the validations it will show the validation based on those values so predictors are populating for each one but you could also manually push the values for these threshold data elements if you're using some external data source if there are no questions maybe we can give 10 more minutes for exercise because there are just two exercises here just one to check from data entry and one from the validation rule analysis and after that we could start with the next session the notifications here so please practice through the learners guide and let us know if there are any questions