In January 2001, the increase in overdose deaths was reversed by an unpredicted and unprecedented reduction in heroin supply that abruptly affected all Australian jurisdictions. The shortage was most marked in New South Wales, the State with the largest heroin market [6,7], where there were increases in price, dramatic decreases in purity at the street level, and reductions in the ease with which injecting drug users reported being able to obtain the drug (see Figures Figures11–3). The abrupt onset of the shortage and the dramatic reduction in overdose deaths prompted first national, then international debate, about the causes and the policy significance of what came to be called the "Australian heroin shortage" (e.g. [8-12]). In this paper we reflect on insights provided by four years' research into the causes, consequences and policy implications of the reduction in heroin supply.

Regular drug injectors reported less frequent heroin use and more frequent injection of other drugs, such as cocaine in NSW, and amphetamines and benzodiazepines in Victoria [7,20,21]. Some of these IDU engaged in riskier forms of injecting and reported more drug-related harms [17,22,23] but there were no increases in deaths related to this other drug use [17,23-26]. We can be less confident about other harms caused by injecting drug use because the health consequences of psychostimulant drugs are not well captured in existing data collection systems. There were, however, significant impacts upon health services and local law enforcement, who began dealing with users exhibiting the behavioural effects of heavy psychostimulant use, and who increasingly reported significant polydrug use problems [17,22,27,28].

Third, the relative contribution of the pre-shortage "glut" and the shortage itself to heroin related harm can be investigated statistically. We conducted a principal component analysis (PCA) of 17 key indicator data series from NSW in order to isolate a small number of uncorrelated principal components that explain the majority of the variance over time in these indicators [32]. PCA is a useful tool for assessing the relative importance of different changes over time, since it sorts the underlying drivers of variance in the indicators (the principal components) according to the magnitude of their effect. The PCA 'loadings' are used to compare variables, with variables with positive loadings contrasted against variables with negative loadings. We conducted PCA with the months of the data sets treated as variables. The first principal component, which explained 47% of the variance across data series over time, clearly contrasted the months before January 2001 with the months after that time. The second principal component (which explained 8% of the variance) contrasted those months in the peak of the 'glut' (1999–2000) with those before and after. These components suggest that: a) the heroin shortage explained more variance in the data series than the 'glut', and b) the shortage and the glut were independent events (since the principal components were uncorrelated).

We concluded that the shortage was probably due to a confluence of factors reflecting the complexity of the heroin market [34]. One of these factors was probably the increased success of high-level Australian drug law enforcement operations conducted nationally and internationally by the Australian Federal Police and Customs (in cooperation with other agencies internationally). These operations removed key individuals directing a small number of highly centralised drug trafficking networks that had supplied large amounts of heroin to Australia, and seized over 1000 kg of heroin in 2000 [34]. Changes in source countries (such as reduced heroin production or increased methamphetamine production) probably also played a role but these did not explain the abrupt onset or the sustained reduction in heroin supply that occurred in Australia at least a year before any shortages were reported in other countries that sourced heroin from these same regions.

Nor do our findings contradict other research evidence that routine heroin seizures have little or no effect on heroin prices or heroin use [41,42]. The scale of seizures in Australia during 2000, for example, comprised 30% of estimated annual heroin consumption [43] compared with 10% in earlier studies of the effects of routine seizures on heroin price and availability. In addition, key persons in the limited number of centralised trafficking networks that controlled the market were arrested in these operations. These two factors probably combined to make Australia a less attractive destination for large scale heroin traffickers. The disruption of these centralised large scale drug distribution networks seems to have produced a return to the methods of importation that were used before the "glut", that is, multiple importations of smaller quantities of heroin using drug couriers and other methods. This has been reflected in the number of recently highly publicised arrests of Australian heroin "mules" in Australia, Indonesia, Hong Kong and Singapore.

The Australian heroin shortage provides good evidence that the integration of supply, demand and harm reduction measures can substantially reduce the harmful effects of injecting heroin use. It would have been very difficult to monitor these changes had we not had the benefit of established strategic early warning systems to document the changes we were able to examine.