Diabetic retinopathy (DR) is one of the most common microvascular complications in diabetes patients as well as one of the most important causes of blindness worldwide. Therefore, DR screening is vital for health care system to prevent blindness. Although, in Thailand, we utilize the percentages of diabetes patients who undergone DR screening program as an indicator of effectiveness in eyes health services, this indicator is not directly reflect the number of diabetes patients who truly go to see retinal specialists for proper management. As a result of the problem mentioned above, this prospective study aimed to compare referral adherence of diabetes patients screened by artificial intelligence (AI) with integrated system group with diabetes patients screened by health care personnel (manual or conventional group). In this study, we screened DR using fundus camera which connected with digital health platform and divided diabetes patients into artificial intelligence (AI) group and manual (conventional) group week by week for 8 month period. However, fundus photo in AI group were reviewed by retinal specialist (overreader) from tertiary hospital and mismatch cases were contacted. Furthermore, referral information was recorded by nurse at tertiary hospital to evaluate referral adherence. Total 1,454 patients included for our analysis (708 patients were screened by AI and 746 patients were screened by manual). Referral adherence of patients diagnosed as referral diabetic retinopathy in AI group was more than manual group and even more than manual group who referred from uncertain fundus images (89.15%, 72.7% and 70.73%, respectively.) Moreover, screening health care personnel’s satisfaction increased after deployment of AI. In conclusion, implementation of an integrated system of artificial intelligence in real clinical setting may improve efficiency in DR screening and also increase referral adherence of diabetic retinopathy patients.