Assessment of patient complexity using routinely collected data: The UK CHIC study

We identified predictors of clinical complexity based on data collected in the UK CHIC Study. All subjects under established care (>1 year) from 2000–2010 were included. Each subject's follow‐up (1 year after study entry to last clinic visit, death or 31/12/2010) was stratified into a series of 6–monthly periods and his/her status was assessed at the start of each. Using Poisson regression (with generalised estimating equations to allow for multiple entries per subject), we studied associations between demographic/clinical factors, CD4 count/percent, viral load (VL), calendar year and measures of prior/current antiretroviral (ART) use, and the development of a new AIDS event or death during each period. A complexity score was derived from the coefficients of the final model; subjects were categorised into ten equally sized groups based on the score, and event rates were calculated for each group. The 31,338 eligible subjects had a median (interquartile range) age of 36 (10, 42) years at baseline. Ethnicity was white (55%), black African (27%), black other (5%), other (9%) and unknown (4%). Mode of acquisition was sex between men (52%), heterosexual sex (37%), other (5%) and unknown (5%). Subjects contributed a total of 377,284 periods of follow‐up (181,170 person‐years [PY]) of which 5796 included a clinical event (rate/1000 PY: 3.20 [95% confidence interval 3.12, 3.28], 4322 AIDS events, 1534 deaths). As an active AIDS‐defining event in the past 6 months was the dominant predictor of a new clinical event (relative rate 41.55), subjects with an active event were excluded from further analysis. Risk factors for a clinical event in patients without an active AIDS event ( Table 1) were earlier calendar year, non‐white ethnicity, older age, lower CD4 count,>80 CD4 cell drop from previous visit, being off ART or on ART with a VL>10,000 copies/ml. 1 Estimates from multivariable Poisson regression model of factors associated with a new clinical event over subsequent 6‐month period (excluding patients with an active AIDS event)RR* (95% CI) P‐valueCalendar year / later year 0.97 (0.96, 0.98) 0.0001Ethnicity– White 1 – – Black African 1.23 (1.13, 1.34) 0.0001 Black other 2.34 (2.06, 2.65) 0.0001 Other 1.64 (1.24, 2.17) 0.0001 Unknown 1.64 (1.38, 1.94) 0.0006Age (years)<30 0.59 (0.51, 0.69) 0.0001 ≥30,<35 0.72 (0.64, 0.80) 0.0001 ≥35,<40 0.75 (0.69, 0.83) 0.0001 ≥40,<45 0.84 (0.76, 0.92) 0.0001 ≥45 1 – –CD4 count (cells/mm 3 )<200 2.49 (2.25, 2.75) 0.0001 200–349 1 – – 350–499 0.72 (0.65, 0.80) 0.0001 ≥500 0.51 (0.46, 0.57) 0.0001 Unknown 0.71 (0.54, 0.93) 0.01 >6 Months of immune suppression 0.84 (0.76, 0.93) 0.001 >80 Cell drop in CD4 count from previous visit 1.15 (1.05, 1.27) 0.004 Previous (non‐active) AIDS event 2.51 (2.31, 2.74) 0.0001Viral load/HAART statusOn HAART VL<1000 c/ml 1 – – On HAART VL≥1000 c/ml 1.81 (1.62, 2.02) 0.0001 On HAART VL≤10000 c/ml 1.23 (1.10, 1.38) 0.0002 On HAART VL>10000 c/ml 2.61 (2.38, 2.86) 0.0001 Missing viral load 2.00 (1.55, 2.56) 0.0001 Hepatitis B/C co‐infection 0.81 (0.73, 0.89) 0.0001 Hepatitis co‐infection and previous experience of immune suppression were associated with lower clinical risk. A score based on this model discriminated reasonably well between subjects who did/did not develop an endpoint over the next 6 months (approximate C‐statistic: 0.72), with event rates increasing from 0.49/100 PY in the lowest score group to 7.16/100 PY in the highest. A score based on clinical markers may provide a means to identify those who will experience clinical progression over the next 6 months, allowing this group to be targeted for closer monitoring and funds for HIV care to be distributed appropriately.