The most frequently studied initiatives to improve community-based services for people with serious mental illness are based on small, integrated, interdisciplinary treatment teams. Endowed with unique resources and staff with specialized skills, programs such as assertive community treatment, supported employment, supported housing, and collaborative treatment of depression are based on the premise that integrated service teams can provide more continuous, more intensive, more collaborative, and ultimately, more effective care.

In this issue, Mark S. Bauer, M.D., and his colleagues report that an extension of this strategy to the treatment of bipolar disorder improved some outcomes, at no increase in total costs. As with other integrated care models, the benefits are modest and do not extend to all major outcome domains but are nevertheless encouraging. Two other articles in this issue—both from the PRISM-E study of outcomes among elderly persons—also report on integrated care models.

Coordinated care models seem to many to be less widely implemented than the evidence base would suggest. Three commonly cited explanations are general resistance to change, the promise of benefits of only moderate magnitude, and high start-up costs. Less attention has been focused on a concern of health system administrators that although, from one perspective, specialized programs may provide more integrated care, from another perspective, they tend to act in relative isolation from larger clinical systems and thus may increase fragmentation.

A paradox of health administration is that "integrated" programs specializing in the treatment of one subgroup of clients not infrequently become proverbial chimneys, silos, or smokestacks walled off from broader health systems. The overall system may thus become more, not less, fragmented and less, not more, capable of addressing clinical complexities. Many clients do not fit into the specialized niches targeted by small integrated teams, and although some colleagues appreciate the effectiveness of "evidence-based" service teams, others find that specialized programs with low caseloads absorb undue resources and are inclined to erect barriers to referral. This phenomenon is not new. The internecine struggles between addictions specialists and general psychiatrists, between surgeons and internists, and even between doctors and lawyers all attest to the thickening of boundaries in response to the development of specialized knowledge.

This paradox deserves further discussion. It may be best addressed through the use of credible performance data pertinent to the specific systems in which new initiatives are implemented. Such data may allow administrators to determine whether or not the benefits of specialized teams justify the concomitant loss in system unity.