Long-term loss to follow-up in the TREAT Asia HIV Observational Database (TAHOD)

Predicting the loss to follow-up (LTFU) of HIV/AIDS patients in China using a recency-frequency (RF) model

BACKGROUND

We constructed a recency-frequency (RF) model for predicting the loss to follow-up (LTFU) in HIV/AIDS patients in China.

METHODS

Data on HIV/AIDS outpatients in the research unit from 1 August 2009 to 30 September 2020 and from 1 October to 31 December 2020 were exported as the observation and prediction datasets, respectively. The classic recency-frequency-monetary (RFM) model was expanded into RFm, RF, RFL and RFmL models. In the observation dataset, the best predictive model was obtained using k-means clustering and C5.0 verification. Then, two rounds of k-means modelling were performed on the best model: data with R ≤ 6 months were retained, randomly divided into a training set (70%) and a testing set (30%) and used to perform the second round of modelling to subdivide the churn and non-churn patients. Next, an ANN algorithm was used to predict LTFU, and the confusion matrix with prediction datasets was constructed.

RESULTS

The observation and prediction datasets included 16 949 and 10 748 samples, respectively. The RF model with three clusters and a quality of 0.82 was the best predictive model. From the observation set, 13 799 samples were retained, and the model accuracy was 100% on the training and testing sets. These 13 799 samples were subdivided into 1563 samples of churn patients and 12 216 samples of non-churn patients. The accuracy of ANN prediction was 99.89%. The accuracy and precision of the confusion matrix were 85.41% and 99.76%, respectively.

CONCLUSION

The RF model is effective in predicting the LTFU of HIV/AIDS patients in China and preventing its occurrence.

Baseline and Process Factors of Anti-Retroviral Therapy That Predict Loss to Follow-up Among People Living with HIV/AIDS in China: A Retrospective Cohort Study

We explored the predictors and predictive models of loss to follow-up (LTFU) during the first year of anti-retroviral therapy (ART). LTFU was defined as the failure to visit the clinic for antiretroviral drugs for ≥ 90 days after the last missed scheduled visit. Based on the electronic medical records of 5953 patients who were HIV positive and began ART between 2016 and 2019 in China, the LTFU rate was 7.24 (95% confidence interval 6.49-7.97) per 100 person-years during the first year of ART. ART baseline factors were associated with LTFU, but were non-optimal predictors. A model including ART process-related factors such as follow-up behaviors and physical health status had an area under the receiver operating characteristic curve of 73.4% for predicting LTFU. Therefore, the medical records of follow-up visits can be used to identify patients with a high risk of LTFU and allow interventions to be implemented proactively.

Trends in Follow-Up Visits Among People Living With HIV: Results From the TREAT Asia and Australian HIV Observational Databases

BACKGROUND

Less frequent follow-up visits may reduce the burden on people living with HIV (PLHIV) and health care facilities. We aimed to assess trends in follow-up visits and survival outcomes among PLHIV in Asia and Australasia.

SETTINGS

PLHIV enrolled in TREAT Asia HIV Observational Database (TAHOD) or Australian HIV Observational Database (AHOD) from 2008 to 2017 were included.

METHODS

Follow-up visits included laboratory testing and clinic visit dates. Visit rates and survival were analyzed using repeated measure Poisson regression and competing risk regression, respectively. Additional analyses were limited to stable PLHIV with viral load <1000 copies/mL and self-reported adherence ≥95%.

RESULTS

We included 7707 PLHIV from TAHOD and 3289 PLHIV from AHOD. Visit rates were 4.33 per person-years (/PYS) in TAHOD and 3.68/PYS in AHOD. Both TAHOD and AHOD showed decreasing visit rates in later calendar years compared with that in years 2008-2009 (P < 0.001 for both cohorts). Compared with PLHIV with 2 visits, those with ≥4 visits had poorer survival: TAHOD ≥4 visits, subhazard ratio (SHR) = 1.88, 95% confidence interval (CI): 1.16 to 3.03, P = 0.010; AHOD ≥4 visits, SHR = 1.80, 95% CI: 1.10 to 2.97, P = 0.020; whereas those with ≤1 visit showed no differences in mortality. The association remained evident among stable PLHIV: TAHOD ≥4 visits, SHR = 5.79, 95% CI: 1.84 to 18.24, P = 0.003; AHOD ≥4 visits, SHR = 2.15, 95% CI: 1.20 to 3.85, P = 0.010, compared with 2 visits.

CONCLUSIONS

Both TAHOD and AHOD visit rates have declined. Less frequent visits did not affect survival outcomes; however, poorer health possibly leads to increased follow-up and higher mortality. Reducing visit frequency may be achievable among PLHIV with no other medical complications.

The impact of data quality and source data verification on epidemiologic inference: a practical application using HIV observational data

BACKGROUND

Data audits are often evaluated soon after completion, even though the identification of systematic issues may lead to additional data quality improvements in the future. In this study, we assess the impact of the entire data audit process on subsequent statistical analyses.

METHODS

We conducted on-site audits of datasets from nine international HIV care sites. Error rates were quantified for key demographic and clinical variables among a subset of records randomly selected for auditing. Based on audit results, some sites were tasked with targeted validation of high-error-rate variables resulting in a post-audit dataset. We estimated the times from antiretroviral therapy initiation until death and first AIDS-defining event using the pre-audit data, the audit data, and the post-audit data.

RESULTS

The overall discrepancy rate between pre-audit and audit data (n = 250) across all audited variables was 17.1%. The estimated probability of mortality and an AIDS-defining event over time was higher in the audited data relative to the pre-audit data. Among patients represented in both the post-audit and pre-audit cohorts (n = 18,999), AIDS and mortality estimates also were higher in the post-audit data.

CONCLUSION

Though some changes may have occurred independently, our findings suggest that improved data quality following the audit may impact epidemiological inferences.