This study compared the predicted data with the actual data captured from the registration system in Jiangsu China during the first three months of COVID-19 measures in Jiangsu, China.
Data source
All actual data were extracted from the web-based Comprehensive Response Information Management System (CRIMS), which is a real-time case reporting system for HIV care services in China. All clients that received positive HIV confirmation test results are linked to the CRIMS and are entitled to free health care services such as physical and mental health care, referral care, CD4 testing, viral load testing once per year, ART services, amongst others. Epidemiological characteristics and clinical results of PLWHIVs collated by health care workers for reporting on CRIMS including patient name, identification card number, age, occupation, route of infection, screening sites, CD4 cell count, ART, and so on.
In total, 2,500 laboratories made up of 14 confirmatory laboratories, 699 screening laboratories, and 1787 screening sites, are designated for HIV testing services in Jiangsu, as at 2019. These facilities reported data on the number of screened or confirmed tests via a web-based system monthly and the testing data used in this study were retrieved from this system.
COVID-19 related case reports for Jiangsu on the other hand were retrieved from publicly available disease databases of Jiangsu provincial Health Committee (http://wjw.jiangsu.gov.cn/) and merged for analysis.
Definition of characteristics
Migratory status was determined using patients’ official household registration details. Participant’s whose registration were not in Jiangsu, was categorized as immigrant [10]. Screening test sites were classified into four groups consisting of: clinical sites, key population sites, penitentiary sites, and others. Data on HIV testing services accessed by persons who sought facility-based healthcare services such as surgery, antenatal care, and sexual transmitted infection testing or medical examination were categorized as hospital sites data. Screening test data that reported from voluntary counseling and testing (VCT), pre-marital examination centers, HIV-positive sexual partners or children testing, physical examination entertainment venue employees, blood donors and conscription physical examination are categorized as key population sites data. Test reports that came from detention and prison centers were categorized as penitentiary sites data whiles testing results obtained via other means not stated were categorized as other site data.
Prediction of the amount of services should be provided during the first three months of COVID-19 measures
The HIV/AIDS and COVID-19 epidemic curves were constructed using data from Jan 1st, 2020 to Mar 31st, 2020 (we define this as the first three months of the COVID-19 measures in China) and key dates relating to epidemic features and control policies were overlaid to aid interpretation (Supplement fig 1). Autoregressive integrated moving average (ARIMA) model was used for forecasting parameters from 2016 to 2020. The normal formular of ARIMA is written as ARIMA (p,d,q)(P,D,Q)s. Four synergistic steps is to developed an ARIMA model, including time series stationary, model identification, parameter estimation and diagnostic checking[17]. First, we choose sequence diagram and white noise test to test whether the time series is stationary or not with log transformation, seasonal or non-seasonal difference. The augmented Dickey-Fuller (ADF) unit root test is carried out to verify the stability of the sequence. Secondly, autocorrelation function (ACF) and partial autocorrelation (PACF) graph were applied to determine the approximate value of p, q, P and Q. Thirdly, the actual ARIMA model was determined by the parametric test and residual test. Box’ test was used to test whether parameters were statistical significance and whether the residual is white noise. Lastly, according to Akaike information criterion (AIC)and Bayesian information criterion (BIC)criteria, the model with the smallest AIC and BIC values is selected as the optimal model [18]. If the ARIMA model does not fit for the data, an alternative model of exponential smoothing model was used for forecasting parameters. All the parameters in the ARIMA model were showed in the supplement materials (Supplement fig 3, Supplement table 1). The error rate is calculated to testified the accuracy of forecast model using data from 2016 to 2019(Supplement fig 4).
Statistical analysis
All curves were forecasted using R packages in R software with version 3.6.3in a time series model for Jan 2020 to Mar 2020 based on the monthly number of reported data from Jan 2016 to Dec 2019. Then we analysis the difference characteristics between the first quarter in 2020 and cumulative numbers in the first quarter for 2016 to 2019 using R software with version 3.6.3. All statistical significance level was set at a=0.05.
Ethical statement
Participants provided written informed consent prior to the interview and blood collection. Information on clients’ personal identifying information such as name, ID number, phone number, and home address were eliminated prior to data analysis. The study process and contents were reviewed and approved by the Institutional Review Board of the National Center for AIDS/STD Control and Prevention (NCAIDS) and Center for Disease Control and Prevention (CDC) in China. The study was carried out in accordance with relevant guidelines.