search

USAID-OHA-SI / TeamTracking

Project Management Tracker for the SI Team
0 stars 0 forks source link

Lesotho: Estimate % loss in TX_CURR due to revised LTFU guidance for FY19 MER 2.3 reporting #62

Open jacobbuehler opened 5 years ago

jacobbuehler commented 5 years ago

Working with the Lesotho country team to determine a "loss factor" that estimates the difference between a TX_CURR cohort inclusive of a 4-week LTFU criteria vs. 90-day LTFU criteria.

We are generating a comparison between FY19 Q1 TX_CURR totals compared to FY18 Q4 totals and time series correcting the Q1 data back to Q4 to determine the % difference.

Will update with more details shortly.

Source data: DATIM FY18 and FY19 unapproved TX_CURR and TX_NEW totals, exported at site level.

jacobbuehler commented 5 years ago

Generated a simplified TX_CURR / TX_NEW x SITE trend over time dataset using DATIM pivot export. DATIM favorite link here (note: FY19Q1 data only available to LS users). After downloading to Excel, I combined DSD/TA totals as separate columns.

Then, to estimate the effect of changing the LTFU guidelines using FY18Q4 and FY19Q1 data, I attempted to "time series correct" the reported Q1 data (28-day LTFU) in order to compare this Q4 estimate with the existing FY18Q4 TX_CURR results (90-day LTFU). Here's how I did this:

*FY19Q1_TX_CURR - (FY19Q1_TX_NEW .9) = estimated "time-corrected" FY18Q4 results assuming 28-day LTFU criteria **

Then, I generated an estimated percentage change due to the LTFU criteria change comparing the estimated Q4 TX_CURR value (28) with the actual Q4 TX_CURR value (90) and divided that difference into the actual value.

Then generated a new column in my pivot data to calculate this value for each site.

Dataset and pivot chart visuals can be reviewed here.

At the national level, the mean % loss is estimated at about 3.7%, whereas the media % loss is estimated at about 2.5%. I haven't visualized this per se--more of a back of the envelope from the pivot data.

The basic visuals I created for the team were TX_CURR trends over time by district or by site, and a comparison between TX_CURR volume and the estimated % loss (a positive value over 3.7% is bad; outliers can be found up to about 30%).

Working with the LS team to determine how this can be better used/interpreted, but it's a starting point to help them explain how they have flatlined treatment results in Q1 and what the estimated effect of the LTFU criteria change could be. Further, the overlay may help the team identify sites where there were significant losses in their TX cohort from Q4 to Q1 beyond the median 2.5% and follow up with the partner to determine if this related to data quality, patient monitoring procedures, etc.