Abstract:
Medical Records are important in patient care, follow-up and clinical research. However medical records often become dormant due to cessation of patient-healthcare provider interaction. Retention of dormant record isinefficient, ineffective, wastes time and resources for storage and may hinder retrieval of active medical records. Knowledge of time-to-dormancy of these records is important to formulate retention and disposal policies for medical records management. However, there is paucity of information on time-to-dormancy of medical records in Nigeria. This study therefore was conducted to determine the statistical distribution, estimates of time-to-dormancy and predictors of record dormancy at the University College Hospital, Ibadan, Nigeria.
A review of medical records from 1990-2014 was conducted at University College Hospital, Ibadan. From 478,300 available records within the study period, systematic sampling technique was used to select 7,685 records. Information on patient’s characteristics (date of first and last visits, gender, age, clinic attended, and other clinical and treatment-outcomes) were extracted from each record using a data extraction proforma. The outcome variablewas time-to-dormancymeasured as the period from creation of a record to the point at which the record becomesdormant. Data analyses were done using descriptive statistics and Kaplan-Meier Method. Estimatedhazard rates of dormancy were plotted against time, log of cumulative hazards[log-log(S(t)]were plotted on log of time (log(t))to determine the statistical distribution and its shape parameter wasestimated. Parametric hazard model was used to identify determinants of time-to-dormancy. Performance of model of choice was compared to a semi-parametric Cox Proportional Hazard(CPH)model. Log likelihood (-2logL)and Akaike Information Criterion (AIC) were used to evaluate CPH and Weibull models that best fitted time-to-dormancy data, while statistical significance was set at .
Patientsage 31-60 years were 40.3%, male constituted 52.4%, and 55.4%resided in Oyo State. Hospital admission rate was 30.0%, while 98.8% patients were alive at the time of last entry. Records with ≥2 entries attained dormancy in 151.9 months (95% CI=128.7-179.1). Hazard plots of time-to-dormancy exhibited a bathtub shape,[log-log(S(t)] on log(t) plots indicated a linear relationship,with estimated shape parameter of 0.6, suggesting Weibull distribution. Values of-2logL forCPH (11061.4) and Weibull (4371.9);and AIC for CPH (11075.4) and Weibull (4389.9). Weibull model indicated that being female (HR=1.1,CI=1.0-1.2); admitted-patient (HR=1.2,CI=1.0-1.4); attendance at Surgical Out-patient (HR=1.1,CI=0.9-1.3); discharged against medical advice (DAMA) (HR=9.0,CI=2.1-36.1) and death (HR=3.6,CI=0.5-25.9), were associated with dormancy. Similarly, CPH regression model indicated that female (HR=1.1,CI=1.0-1.3); admitted-patient (HR=1.2,CI=1.0-1.4); attendance at Surgical Out-patient (HR=1.0,CI=0.9-1.3); DAMA(HR=17.9,CI=4.3-74.9) and death (HR=3.1,CI=0.4-22.4), equally influenced dormancy.
Weibull model provided the best fit suggesting a minimum retention period of 151.9 months for medical records.Records of females, admitted-patients,those who attended surgical out-patient, patients discharged against medical advice and deadpatientsare more likely to become dormant earlier. A medical records retention policy should be formulated based on the estimated time-to-dormancy.
