We used risk adjusted statistical process control methods^w5 to follow the changes in death rates at Walsall Hospitals NHS Trust from February 1996 to September 2004 (figure). The chart crosses the alarm threshold many times between 1996 and 2001 (implying that the risk of death at Walsall often rose to at least twice the national average during that period) but after early 2002 crossed only twice. With indirect standardisation, the HSMR at Walsall dropped from 130.0 in 2000 to 92.8 by September 2004 (see figure on bmj.com). For the diagnoses leading to 80% of all deaths, the overall reduction in the number of observed minus expected deaths was 295 (27% of observed deaths in 2001). The largest reductions were in circulatory diseases (44%) and respiratory diseases (30%). This works out at 0.14 fewer observed minus expected deaths per bed for 100% of diagnoses per year. Part of the reduction in HSMR could be due to chance, regression to the mean, better coding, different discharge policies, or referral of more complicated cases to other hospitals but, as the reductions coincided with an extensive effort to implement improvement policies, they are probably associated with improved quality of care. The number of hospital doctors per bed is the best predictor of HSMRs (lower HSMRs being associated with more doctors per bed)^w1: from 1999 to 2003 Walsall's ratio of doctors per bed increased from 0.226 to 0.305.